Modellierung des Stickstoffhaushaltes für den

Transcrição

Modellierung des Stickstoffhaushaltes
für den Gewässerschutz mittelgroßer
Flusseinzugsgebiete
Kumulativ-Dissertation
Justus-Liebig-Universität Gießen
vorgelegt durch:
Dipl.-Umweltwiss. Thorsten Pohlert
Gießen, den 19.12.2006
Danksagung
Dieses Werk wäre ohne die Unterstützung zahlreicher Menschen nicht möglich gewesen;
bei diesen möchte ich mich an dieser Stelle bedanken. Bei Prof. Dr. Hans-Georg Frede
möchte ich mich für die Betreuung meiner Doktorarbeit und das entgegengebrachte
Vertrauen bedanken. Herrn Prof. Dr. King danke ich für die Übernahme dieser Arbeit
als Gutachter. Für die wertvollen wissenschaftlichen Diskussionen und Tipps bei der
Publikation von Zeitschriftenartikeln möchte ich mich bei Dr. Lutz Breuer und Dr. Johan
A. Huisman bedanken. Herzlich bedanken möchte ich mich bei Beate Lindenstruth, Nelly
Weiß und Adrian Skitschak für die wichtige Mithilfe in der Betreuung der Messstelle
Aßlar, der Mithilfe bei den Längsprofilmessungen und die fließige Arbeit im Labor. Bei
Dr. Marin Bach und Oliver Ginzler möchte ich mich für die gute Zusammenarbeit der
letzten Projekte bedanken. Wolfgang Reiher gebührt Dank für seine Geduld, mit mir bei
griechischem Gyros über LINUX und Programmierung zu fachsimpeln. Bei sämtlichen
Mitarbeiterinnen und Mitarbeitern im Institut für Ressourcenmanagement möchte ich
mich für das äußerst angenehme Arbeitsklima bedanken. Es war ein sehr schöne Zeit mit
Euch zusammenzuarbeiten und ich werde Euch bald vermissen.
Meine Promotion wurde im Rahmen des von der Deutschen Forschungsgemeinschaft
unterstützten Sonderforschungsbereichs 299 “Landnutzungskonzepte für periphere Regionen (SFB299)” an der Justus-Liebig-Universität Gießen angefertigt. Für die finanzielle
Förderung des Forschungsprojekts durch die DFG sei an dieser Stelle gedankt.
Dank an meine Eltern Ingeborg und Harald Pohlert sowie an meine Großmutter
Elisabeth Pohlert für die langjährige Unterstützung meines Studiums und der Promotion.
Bei meiner Frau Judith Lehnart möchte ich mich für den liebevollen Rückhalt und das
entgegengebrachte Verständnis bedanken.
iii
Danksagung
Für Nathanael
iv
Inhaltsverzeichnis
Danksagung
iii
1 Erweiterte Zusammenfassung
1.1 Einleitung . . . . . . . . . . . . . . . . . . . .
1.2 Auswahl eines Computermodells . . . . . . . .
1.3 Aufbau des Einzugsgebietsmodells . . . . . . .
1.3.1 Das Dill Einzugsgebiet . . . . . . . . .
1.3.2 Geobasisdaten . . . . . . . . . . . . . .
1.3.3 Modellinput . . . . . . . . . . . . . . .
1.3.4 Vergleichsgrößen . . . . . . . . . . . .
1.4 Test und Ablehnung des Modells . . . . . . .
1.4.1 Ergebnisse und Diskussion . . . . . . .
1.4.2 Schlussfolgerung . . . . . . . . . . . . .
1.5 Formulierung und Test eines neuen Modells .
1.5.1 Das Modell SWAT-N . . . . . . . . . .
1.5.2 Sensitivitäts- und Unsicherheitsanalyse
1.5.3 Validierung von SWAT-N . . . . . . .
1.5.4 Schlussfolgerung . . . . . . . . . . . . .
1.6 Anwendung des neuen Modells . . . . . . . . .
1.7 Fazit und Ausblick . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
3
5
5
5
7
9
10
10
13
13
13
15
17
21
21
23
.
.
.
.
.
.
.
.
.
.
25
25
26
26
27
27
28
28
29
33
33
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 Pohlert et al., 2005, Adv. Geosci., 5, 7–12.
2.1 Introduction . . . . . . . . . . . . . . . . . . . .
2.2 Materials and Methods . . . . . . . . . . . . . .
2.2.1 The Dill catchment . . . . . . . . . . . .
2.2.2 The SWAT model . . . . . . . . . . . . .
2.2.3 Input data . . . . . . . . . . . . . . . . .
2.2.4 Discharge and nitrate load measurements
2.2.5 Calibration . . . . . . . . . . . . . . . .
2.3 Results and Discussion . . . . . . . . . . . . . .
2.4 Conclusions . . . . . . . . . . . . . . . . . . . .
2.5 Acknowledgements . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
v
Inhaltsverzeichnis
3 Pohlert et al., 2006, Ecol. Model., in press.
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Model description . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 N input into the soil system . . . . . . . . . . . . .
3.2.2 Allocation of organic matter to organic litter pools
3.2.3 Decomposition of soil organic matter . . . . . . . .
3.2.4 Clay adsorption . . . . . . . . . . . . . . . . . . . .
3.2.5 Ammonia volatilisation . . . . . . . . . . . . . . . .
3.2.6 Nitrification . . . . . . . . . . . . . . . . . . . . . .
3.2.7 Denitrification . . . . . . . . . . . . . . . . . . . . .
3.2.8 Plant uptake of N and N fixation . . . . . . . . . .
3.2.9 N transport . . . . . . . . . . . . . . . . . . . . . .
3.3 Data set for model testing . . . . . . . . . . . . . . . . . .
3.4 Model calibration . . . . . . . . . . . . . . . . . . . . . . .
3.5 Sensitivity analysis . . . . . . . . . . . . . . . . . . . . . .
3.5.1 Regression based global sensitivity analysis . . . . .
3.5.2 Results of the sensitivity analysis . . . . . . . . . .
3.6 Generalised likelihood uncertainty estimation (GLUE) . . .
3.6.1 Results of the GLUE analysis . . . . . . . . . . . .
3.6.2 Model validation . . . . . . . . . . . . . . . . . . .
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.8 Acknowledgements . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Pohlert et al., 2006, Hydrol. Earth Syst. Sci. Discuss.,3, 2813–2851.
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 The Dill catchment . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 Spatial input data . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 Nitrogen input data . . . . . . . . . . . . . . . . . . . . . . .
4.2.4 Hydrologic and meteorological data . . . . . . . . . . . . . .
4.2.5 Hydrochemical Data . . . . . . . . . . . . . . . . . . . . . .
4.2.6 The SWAT-N model . . . . . . . . . . . . . . . . . . . . . .
4.2.7 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.8 Assessment of model performance . . . . . . . . . . . . . . .
4.3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Discharge . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2 Nitrate loads . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Longitudinal profile of nitrate concentrations . . . . . . . . .
4.3.4 Geographical sources of nitrate leaching . . . . . . . . . . .
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
36
37
39
40
40
42
42
42
43
44
45
45
46
47
47
50
52
52
53
53
56
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
59
59
63
63
64
64
64
65
66
67
68
69
69
72
75
76
78
79
Inhaltsverzeichnis
A Hilfsprogramme für SWAT-N
A.1 A short manual for SWAT-N . . . . . . . . . . . . . . . . . .
A.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . .
A.1.2 How the program works . . . . . . . . . . . . . . . .
A.1.3 Pre-processing . . . . . . . . . . . . . . . . . . . . . .
A.1.4 Post-processing . . . . . . . . . . . . . . . . . . . . .
A.1.5 Note for programmers . . . . . . . . . . . . . . . . .
A.2 addph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2.1 Einleitung . . . . . . . . . . . . . . . . . . . . . . . .
A.2.2 Was ist das Problem? . . . . . . . . . . . . . . . . . .
A.2.3 Was kann das Programm? . . . . . . . . . . . . . . .
A.2.4 Benutzung von AddpH . . . . . . . . . . . . . . . . .
A.2.5 Ergebnis . . . . . . . . . . . . . . . . . . . . . . . . .
A.2.6 Limitierungen . . . . . . . . . . . . . . . . . . . . . .
A.3 idw – Inverse Distant Weighted Average . . . . . . . . . . .
A.3.1 Ziel . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.3.2 Funktionsweise . . . . . . . . . . . . . . . . . . . . .
A.3.3 Wichtige Hinweise . . . . . . . . . . . . . . . . . . .
A.4 READOUT . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.4.1 What is READOUT? . . . . . . . . . . . . . . . . . .
A.4.2 Installation . . . . . . . . . . . . . . . . . . . . . . .
A.4.3 Starting READOUT . . . . . . . . . . . . . . . . . .
A.4.4 Warning . . . . . . . . . . . . . . . . . . . . . . . . .
A.4.5 Importing READOUT results into other applications
A.4.6 Error messages and limitations . . . . . . . . . . . .
A.4.7 De-installation . . . . . . . . . . . . . . . . . . . . .
A.4.8 To do . . . . . . . . . . . . . . . . . . . . . . . . . .
A.4.9 License (short) . . . . . . . . . . . . . . . . . . . . .
A.4.10 Disclaimer (short) . . . . . . . . . . . . . . . . . . . .
A.5 GoF – Goodness-of-Fit . . . . . . . . . . . . . . . . . . . . .
A.5.1 Funktionalität . . . . . . . . . . . . . . . . . . . . . .
A.5.2 Anwendung . . . . . . . . . . . . . . . . . . . . . . .
A.6 Manual for hru2map . . . . . . . . . . . . . . . . . . . . . .
A.6.1 Purpose of hru2map . . . . . . . . . . . . . . . . . .
A.6.2 History of hru2map . . . . . . . . . . . . . . . . . . .
A.6.3 How the program works . . . . . . . . . . . . . . . .
A.6.4 How to use the program . . . . . . . . . . . . . . . .
A.6.5 Technical limitations . . . . . . . . . . . . . . . . . .
A.6.6 Information for programmers . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
81
82
82
82
84
85
85
85
85
85
86
87
89
90
90
90
91
91
92
92
92
93
94
94
96
96
97
97
97
97
97
98
98
98
98
100
100
102
102
vii
Inhaltsverzeichnis
viii
Abbildungsverzeichnis
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2.1
2.2
2.3
2.4
2.5
3.1
3.2
Modellierung von Umweltsystemen (Refsgaard & Henriksen, 2004, S. 75). . .
Alter Ansatz zur Beschreibung des N-Kreislaufs in SWAT-G (Pohlert et al.,
2006a, siehe auch: Kapitel 3, S. 38). . . . . . . . . . . . . . . . . . . . . . .
Das Einzugsgebiet der Dill (Pohlert et al., 2006b, siehe auch Kapitel 4, S. 65).
Simulierte und beobachtete monatliche Nitratfracht (kg N d−1 ) mit SWAT-G
für die Periode April 2000 bis Januar 2003 am Pegel Aßlar (Pohlert et al., 2005b,
siehe auch Kapitel 2, S. 30) . . . . . . . . . . . . . . . . . . . . . . . . . .
Mit SWAT-G simulierte (grau) und beobachtete (schwarz) Nitratauswaschung
des Lysimeters in Brandis, Sachsen (Pohlert et al., 2006a, siehe auch Kapitel 3,
S. 55) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Neuer Ansatz zur Beschreibung des N-Kreislaufs in SWAT-N (Pohlert et al.,
2006a, siehe auch Kapitel 3, S. 39) . . . . . . . . . . . . . . . . . . . . . . .
Mit SWAT-N simulierte (grau) und beobachtete (schwarz) Nitratauswaschung
S. 55) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Vergleich zwischen simulierten (rot) und beobachteten (weiß) Nitratkonzentrationen diverser Messstellen entlang der Dill (links) und deren Zuflüsse
(rechts)(vgl. Kapitel 4, S. 77). Die Lage der Messstellen ist Abb. 4.2, S. 65 zu
entnehmen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Absolute monatliche Differenz der spezifischen Nitrafracht für das AGENDA2000 und CAP Szenario (Pohlert et al., 2005a, S. 10). . . . . . . . . . . .
Stream network and locations of gauged subbasins within the Dill catchment.
Simulated and observed daily discharge [m3 s−1 ] for the validation period (April
2000 to January 2003) at the gauge Asslar. E = 0.56, b = 0.81, R 2 = 0.60. .
Simulated and observed monthly nitrate load [kg N d−1 ] for the period April
2000 to January 2003 at the gauge Asslar. Calibration: E = 0.66, b = 1.26, R 2
= 0.69. Validation: E = 0.70, b = 1.21, R2 = 0.77. . . . . . . . . . . . . . .
Monthly contribution of point sources as a fraction of observed and simulated
total nitrate load at the gauge Asslar. . . . . . . . . . . . . . . . . . . . . .
Denitrification and plant nitrate uptake of winter barley after fertilisation with
80 kg N ha−1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
2
.
.
4
6
. 11
. 12
. 14
. 17
. 20
. 22
. 26
. 29
. 30
. 31
. 32
Original conceptualisation of the N cycle in SWAT . . . . . . . . . . . . . . .
New conceptualisation of the N cycle in SWAT-N. . . . . . . . . . . . . . . .
38
39
ix
Abbildungsverzeichnis
3.3
3.4
3.5
3.6
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
A.1
x
Observed and simulated monthly evapotranspiration at lysimeter station Brandis, Germany (Data taken from Haferkorn, 2000). . . . . . . . . . . . . . . .
Observed and simulated monthly percolation at lysimeter station Brandis, Germany (Data taken from Haferkorn, 2000). . . . . . . . . . . . . . . . . . . .
Dotty plots according to the GLUE methodology applied to six sensitive parameters of SWAT-N focusing nitrogen leaching. The labels of the x-axis provide the
parameter range of the individual parameters. Sensitive parameters were derived
after a global sensitivity analysis. NSE means Nash & Sutcliffe efficiency; for the
definition of other parameters refer to Table 3.6. Please note the logarithmic
scale of k1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A: Results of observed monthly N-leaching (arithmetic mean of three replicates)
at lysimeter station Brandis, Germany, (Haferkorn, 2000) and conditional 95 %
uncertainty interval according to the GLUE methodology simulated with SWATN. B: Results of observed monthly N-leaching and best simulation run of the
previous SWAT version. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Map of actual land-use in the Dill catchment based on LANDSAT TM5 data
(Nöhles, 2000). AGRL means arable land, PAST is pasture, CVEG is changing
vegetation, FRSD is deciduous forest, FRSE is evergreen forest, URBN means
urban area, and WATR denotes surface water. . . . . . . . . . . . . . . . . .
Map of the Dill catchment and its gauged subbasins (left). Locations for
snapshot-sampling along the river Dill and its tributaries(right). . . . . . . .
Scatterplots of simulated and observed discharge at gauges Obere Dill and
Dillenburg. Daily discharge (left) and log-transformed daily discharge (right)
are depicted for the period 1.7.2000 - 31.12.2004. The dashed line indicates the
one-to-one line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Map of simulated annual average groundwater contribution to the reach of each
subbasin (2000-2004). . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulated and observed nitrate load at gauge Aßlar (1.7.2000 - 31.12.2004). .
Simulated and observed daily nitrate load at gauge Obere Dill (1.1.2000 28.2.2003). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison of observed (white) and simulated (red) nitrate concentrations at
various sampling locations along the river channel Dill (left) and at various
tributaries (right) on four different dates. The sampling sites are given in Fig.
4.2 b. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulated average annual nitrate leaching through lateral flow (kg N ha−1 ) for
2000 to 2004 (left). Ratio between nitrate leaching through lateral flow and total
nitrate leaching (%) for the same period (right). . . . . . . . . . . . . . . . .
. 48
. 49
. 54
. 55
. 63
. 65
. 71
. 72
. 73
. 75
. 77
. 78
Example map of re-assigned HRUs for the Dill catchment. White areas contain
no-data values as well as not re-assigned grid cells. . . . . . . . . . . . . . . .
99
Tabellenverzeichnis
1.1
1.2
1.3
1.4
1.5
1.6
1.7
2.1
2.2
2.3
Allgemeine Charakteristika des Einzugsgebiets der Dill (Pohlert et al., 2005b,
siehe auch Kapitel 2, S. 27). . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
Mittlere Jahresflüsse von N für ein typisches, ackerbaulich genutztes HRU im
Einzugsgebietsmodell der Dill simuliert mit SWAT-G. (Pohlert et al., 2005b,
siehe auch Kapitel 2, S. 32) . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
Rangplätze der sensitiven Parameter für die jährliche N-Emission (NE ) und
jährliche N-Auswaschung (NL ). Es sind R2 das Bestimmtheismaß, bSD der standardisierte Regressionskoeffizient und bO der unstandardisierte Regressionskoeffizient (Pohlert et al., 2006a, siehe auch Kapitel 3, Tabelle 3.5). . . . . . . . . .
16
Kalibrierte Parameter des Einzugsgebietsmodells der Dill (1.7.2000 - 31.1.2003).
Die soil-unit 2458 stellt eine flachgründige Braunerde und Soil-unit 202 einen
Auenboden dar. Die räumlich verteilten Parameter wurden nach Eckhardt &
Arnold (2001) kalibriert (siehe auch Kapitel 4, S. 69). . . . . . . . . . . . . . .
18
Modelleffizienzen für simulierten und beobachteten täglichen Abfluss an verschiedenen Pegeln des Dilleinzugsgebiets (Kalibrierung: 1.7.2000 - 31.1.2003,
Validierung: 1.2.2003 - 31.12.2004). E bedeutet Modelleffizienz nach Nash &
Sutcliffe (1970), die Indices d, log, w, und m stehen für tägliche, log-Tageswerte,
wöchentliche und monatliche Werte (siehe auch Kapitel 4, S. 70). . . . . . . . .
18
Modelleffizienzen für die Nitratfracht (1.7.2000-31.12.2004) an den Stationen
Aßlar und Obere Dill. E bedeutet Modelleffizienz nach Nash & Sutcliffe (1970), D
ist der Index of Agreement (Wilmott et al., 1985), N̄o ist die mittlere beobachtete
Nitratfracht (kg N d−1 ), N̄s ist die mittlere simulierte Nitratfracht (kg N d−1 ),
sdo ist die Standardabweichung der beobachteten Werte (kg N d−1 ), sds ist die
Standardabweichung der simulierten Werte (kg N d−1 ) und n gibt die Anzahl
der Datenpaare wieder (vgl. auch Kapitel 4, S. 74). . . . . . . . . . . . . . . .
19
Landnutzungsverteilung (%) des Dilleinzugsgebiets für die Referenz, das
AGENDA2000- und das CAP-Szenario (Pohlert et al., 2005a, S. 9). . . . . . . .
21
Characteristics of the Dill catchment. Information on topography is derived
from the DEM, land use distribution is taken from processed LANDSAT TM5
scenes (Nöhles, 2000) and climate data are taken from the hydrologic yearbook.
27
Manually calibrated parameters which control both nitrogen cycle and transport. 28
Average annual values for the main nitrate fluxes on an example HRU with
agricultural land use. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
xi
Tabellenverzeichnis
3.1
3.2
3.3
3.4
3.5
3.6
3.7
4.1
4.2
4.3
A.1
A.2
A.3
A.4
xii
Soil physical parameters of lysimeter group 5 (Haplic Arenosol) at Brandis
(after Haferkorn, 2000). z is depth of layer, CL is clay content, SL is silt content,
HU M means humus content, KSAT is saturated hydraulic conductivity, and BD
is wet bulk density. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Crop rotation, precipitation (P ), actual evapotranspiration (ET ), N-fertilisation
(NF ), N-deposition (ND ), N-leaching (NL ), and N-uptake by crops (NUP ) for the
lysimeter in Brandis (after Haferkorn, 2000; Dreyhaupt, 2000). Both P and E T
refer to the period November of the preceding year to October of the specified
year, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Initial and calibrated parameter values. Parameters of the top soil layer were
calibrated and the parameters of underlying soil layers were adjusted according
to the change ratio of the top soil layer as proposed by Eckhardt & Arnold (2001).
Quantiles of analysed output variables. Annual gaseous nitrogen emission is
NE , annual nitrogen leaching is NL , and nitrogen crop uptake is NUP , average
daily organic nitrogen content is NORG and mineral nitrogen content is NMIN
(n = 4, 000). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ranking of sensitive parameters of various output variables. R 2 denotes coefficient of determination, bSD is standardized regression coefficient, bO means
ordinary regression coefficient and Rank means absolute numerical importance
(N = 4, 000). For the definition of other symbols refer to the text as well as
Table 3.6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Parameters and their range for the employed sensitivity analysis. . . . . . . . .
Values of parameters which remained fixed for the GSA and GLUE analysis as
well as their sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Calibrated parameters for the Dill catchment (1.7.2000 - 31.1.2003). The soilunit 2458 refers to a shallow soil that is present on both upper slopes and the top
of the hills. Soil-unit 202 is typical for lower slope positions. Spatially distributed
parameters were calibrated as proposed by Eckhardt & Arnold (2001). . . . . .
Goodness-of-fit measures for simulated and observed discharge at various gauges
for the calibration period (1.7.2000 - 31.1.2003) and assessment period (1.2.2003 31.12.2004). E denotes model efficiency after Nash & Sutcliffe (1970), the indices
d, log, w, and m denote daily data, log-transformed data, weekly averages, and
monthly averages, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . .
Goodness-of-fit measures for nitrate load predictions (1.7.2000-31.12.2004) at
gauges Aßlar and Obere Dill. E means model efficiency (Nash & Sutcliffe, 1970),
D is the index of agreement (Wilmott et al., 1985), N̄ is average nitrate load (kg
N d−1 ), sd is standard deviation (kg N d−1 ), n is the number of data pairs, and
the indices o and p mean observed and simulated, respectively. Only monthly
instantaneous samples were available for the Obere Dill. . . . . . . . . . . . . .
Hilfsprogramme für die Modellierung mit SWAT-N . .
New and modified source files included in SWAT-G-N
Liste der Programmdateien zu idw. . . . . . . . . . .
Input files for hru2map. . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
46
47
48
50
51
57
58
69
70
74
81
86
90
101
Tabellenverzeichnis
A.5
A.6
Procedures of hru2map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Modules and their globally defined variables. . . . . . . . . . . . . . . . . . . 104
xiii
Tabellenverzeichnis
xiv
1.1 Einleitung
Die Bedeutung von Stickstoff für die Güte von Oberflächen- und Grundwässern sowie
der Bedarf, Landnutzungsänderungen im Vorfeld auf die Wirkung der Gewässergüte
abschätzen zu können, hat durch die Verabschiedung der Europäischen Wasserrahmenrichtlinie (WRRL, 2000/60/EG) an Bedeutung gewonnen. Zur Beantwortung der Frage,
welche Änderungen in der Landnutzung zu welchen mengenmäßigen Änderungen in der
Stofffracht führen, bietet sich die Anwendung prozessorientierter Simulationsmodelle an.
Die Notwendigkeit der Entwicklung von prozessorientierten Simulationsmodellen für
das Management der durch menschliche Aktivitäten weltweit unter Druck geratenen Ressource Wasser hinsichtlich Verfügbarkeit und Güte wurde durch die PUB-Initiative (Predictions in Ungauged Basins) für die aktuelle Dekade (2003-2013) der IAHS (International
Association of Hydrological Sciences) besonders hervorgehoben (Sivapalan et al., 2003).
Ziel dieser Arbeit ist, ein szenarienfähiges Simulationsmodell für die Prognose von Stickstofffrachten in Abhängigkeit von der Landnutzung im Fließgewässer zu entwickeln und
anzuwenden. Die Anzahl der gewässerchemischen Messstellen ist deutlich geringer als die
Zahl der Pegel, weshalb in der Praxis nur wenige, langjährige Reihen zur N-Fracht für
die Kalibrierung von Modellen zur Verfügung stehen. Daher wird ein Simulationsmodell
benötigt, das auch ohne Kalibrierung der stickstoffbestimmenden Parameter akzeptable
Ergebnisse liefert.
Im Rahmen des Sonderforschungsbereichs “Landnutzungskonzepte für periphere Regionen” (SFB 299) an der Justus-Liebig-Universität Gießen wurde im Jahr 2000 von Lenhart
(2003) ein zeitlich hoch aufgelöstes gewässerchemisches Messprogramm begonnen und es
wurden erste Versuche zur Simulation von N-Frachten unternommen. Das Messprogramm
wurde im Rahmen der vorliegenden Arbeit weitergeführt. Die Ergebnisse dieses Messprogramms sowie vier Längsprofilmessungen zur räumlichen Konsistenzprüfung der Modellergebnisse bilden die wesentliche, eigenständig erhobene Datengrundlage für die Entwicklung, Kalibrierung und Validierung eines Simulationsmodells zur N-Frachtberechnung
mesoskaliger Einzugsgebiete.
Ein Simulationsmodell ist eine vereinfachte mathematische Abbildung eines natürlichen
Systems (hier: Flusseinzugsgebiet), das es erlaubt, rechnergestützte Studien an diesem
Stellvertreterobjekt durchzuführen. Der Aufbau der vorliegenden Kumulativ-Dissertation
ist an die von Refsgaard (1997) vorgeschlagene Arbeitsweise für die Modellierung von
Umweltsystemen angelehnt. In Abb. 1.1 sind der Ablauf und die Grundbegriffe zur Modellierung dargestellt. Nach Refsgaard (1997) gilt es zunächst, den Zweck einer Modellanwendung zu bestimmen, der sich durch die oben angegebene Zielsetzung ergibt. Dann ist
zu überprüfen, ob ein geeignetes Computermodell (Model Code), das auf ein vorhandenes
mathematisches Konzept (Conceptual Model, Abb. 1.1) aufbaut, zur Beantwortung der
1
Abbildung 1.1: Modellierung von Umweltsystemen (Refsgaard & Henriksen, 2004, S. 75).
Fragestellung vorhanden ist. Die Auswahl des Computermodells ist im Abschnitt 1.2 dargestellt. Nach der Wahl eines vorhanden Computermodells wird an Hand von Datensätzen
eines konkreten Einzugsgebiets ein Einzugsgebietsmodell aufgebaut (Model Set-Up). Die
Beschreibung zum Aufbau des Datensatzes für das Einzugsgebietsmodell befindet sich im
Abschnitt 1.3. Das Einzugsgebietsmodell wird mittels Kalibrierung (Model Calibration)
innerhalb physikalisch vertretbarer Grenzen so angepasst, dass die Abweichungen zwischen
Modellausgaben und Systembeobachtungen (Hydrographen, Chemographen) minimiert
werden. Im Anschluss an die Kalibrierung erfolgt die Validierung, d.h. der Vergleich zwischen Modellausgabe und Systembeobachtungen für eine unabhängige Zeitperiode (Model
Validation, Abb. 1.1). Nachdem durch die Validierung des Modells eine Aussage über die
Genauigkeit des Modells getroffen werden kann, erfolgt als Post-Audit die weitere Prüfung
hinsichtlich der Richtigkeit der Modellergebnisse. Dieser als Analysis beschriebene Teil der
Modellierung (Abb. 1.1) wird in Abschnitt 1.4 erläutert.
Lässt sich die Richtigkeit des Modells nicht widerlegen, so kann das Modell angenommen
werden (Model Confirmation). Bestehen jedoch Zweifel an der Richtigkeit der Modellergebnisse, muss in Anlehnung an das Falsifikationsprinzip das Modell verworfen werden
(Model Rejection), was im Abschnitt 1.4 diskutiert wird. Nach der Ablehnung eines Modells folgt, dass der Prozess der Modellierung von neuem beginnt, was durch die Darstellung als Kreislauf zum Ausdruck gebracht wird (Abb. 1.1). Im Abschnitt 1.5 wird ein
neues mathematisches Modell (Conceptual Model ) zur Abbildung des N-Kreislaufs vorgestellt, das den alten Ansatz im bestehenden Computermodell ersetzt. Die Prüfung des
neuen Modellansatzes wird ebenfalls im Abschnitt 1.5 vorgestellt. Im Abschnitt 1.6 wird
ein Anwendungsbeispiel des neuen Modells im Rahmen des SFB299 vorgestellt und in
Abschnitt 1.7 ist ein Forschungsausblick gegeben.
2
1.2 Auswahl eines Computermodells
1.2 Auswahl eines Computermodells
In der letzten Dekade sind durch die hydrologischen Wissenschaften zahlreiche ökohydrologische Modelle entwickelt worden, deren Einsatz für die Berechnung der N-Fracht
als prinzipiell verwendbar erscheint. Es existieren unter anderem einfache Bilanzmodelle,
wie das Modell N-BILANZ (Gebel, 2000), das basierend auf dem jährlichen N-Überschuss
landwirtschaftlicher Flächen den jährlichen N-Eintrag in das Grundwasser berechnet. Ferner gibt es die für große Flusseinzugsgebiete (AEO > 10000 km2 ) entwickelten empirischen Modelle MONERIS (Behrendt et al., 2000) und PolFlow (de Wit, 2001), die mittels
Exportkoeffizienten und Einwohnergleichwerten den jährlichen Anteil der N-Fracht aus
diffusen und punktuellen Quellen abschätzen. Ein Nachteil dieser Modelle ist die eingeschränkte Szenarienfähigkeit, da diese Modelle die Ursache-Wirkungsbeziehungen stark
vereinfachen und die empirischen Koeffizienten der Modelle unter veränderten Landnutzungsbedingungen keine Gültigkeit haben.
Die prozessorientierten Modelle zur Simulation von N-Austrägen aus Einzugsgebieten
lassen sich gemäß des Aggregationsgrads der Gebietseigenschaften von räumlich verteilten,
semi-verteilten, bis hin zu aggregierten Modellen unterteilen. Zur Gruppe der räumlich
verteilten Modelle gehören das MIKE SHE (Refsgaard & Storm, 1995) und das TNT Model (Beaujouan et al., 2001), das auf den hydrologischen Modellkomponenten von TOPMODEL (Beven et al., 1995) aufbaut. Zwar bieten die genannten Modelle ein Höchstmaß
an Komplexität und Flexibilität, was die Abbildung räumlich verteilter Prozesse anbelangt, ein Nachteil dieser Modelle ist jedoch der hohe Rechen- und Datenaufwand. Aus diesen Gründen können die genannten Modelle oft nur für die Simulation kleiner (AEO < 10
km2 ), sehr gut untersuchter Einzugsgebiete unter Zuhilfenahme von Großrechnern angewandt werden.
Einen Kompromiss zwischen Komplexität, Datenverfügbarkeit und Rechenaufwand
stellen semiverteilte Modelle dar. Hierunter fallen das LASCAM (Sivapalan et al., 2002;
Viney et al., 2000) und das HBV-N Modell (Bergström, 1995; Arheimer & Brandt, 1998),
die beide für große Einzugsgebiete (AEO > 10000 km2 ) entwickelt wurden und deren
kleinste zu simulierende Einheit ein topographisches Teileinzugsgebiet darstellt. Im Modell INCA (Whitehead et al., 1998b; Wade et al., 2002) werden die Teileinzugsgebiete
weiter in Flächen gleicher Eigenschaften (Hydrotope, bzw. Hydrological Response Units)
unterteilt. Jedoch benötigt INCA ein weiteres Modell, das die notwendigen hydrologischen
Eingangsdaten berechnet. Ferner wird in INCA der N-Haushalt als offenes System modelliert, da im Modell der organische N-Speicher als unendlich groß aufgefasst wird (Wade
et al., 2002, S. 562). Damit kann mit INCA der Prozess der Mineralisation nur vereinfacht
und der Prozess der Immobilisierung gar nicht simuliert werden. Eine weitergehende Diskussion zum Stand der vorhandenen Stickstoffmodelle sind bei Pohlert et al. (2006a) und
Pohlert et al. (2006b) in den jeweiligen Einleitungsteilen zu finden. Die Fachartikel sind
dieser Kumulativ-Dissertation in den Anhängen 3 und 4 beigefügt.
Für die vorliegende Arbeit wurde eine für periglazial geprägte Mittelgebirgsregionen
modifizierte Version SWAT-G (Eckhardt et al., 2002), des international weit verbreiteten, semiverteilten Modells Soil and Water Assessment Tool (SWAT, Arnold et al.,
1998) verwendet. Der Landschaftswasserhaushalt wird in SWAT-G als zeitliches Kontinuum modelliert, wobei der Rechenzeitschritt einen Tag beträgt. SWAT-G wurde für diese
3
Abbildung 1.2: Alter Ansatz zur Beschreibung des N-Kreislaufs in SWAT-G (Pohlert et al.,
2006a, siehe auch: Kapitel 3, S. 38).
Arbeit gewählt, da hier ein Pflanzenwachstumsmodell integriert ist, das die Simulation
unterschiedlicher Fruchtfolgen innerhalb eines Einzugsgebiets erlaubt. Ferner ist die Modellkonzeption vollständig dokumentiert (Neitsch et al., 2002), und das Programm ist
nach dem amerikanischen Recht als Public Domain veröffentlicht, womit nicht nur das
Programm kostenlos sondern auch der Quellcode frei zugänglich ist. Weiterhin besitzt das
Institut für Ressourcenmanagement mehrjährige Forschungserfahrung mit SWAT-G (z.
B. Haverkamp, 2000; Eckhardt et al., 2002; Lenhart, 2003; Breuer et al., 2003), an die
durch diese Arbeit angeknüpft und die Forschung weitergeführt werden kann.
Abbildung 1.2 stellt schematisch den alten Ansatz zur Modellierung des N-Haushaltes
dar, so wie er in SWAT-G implementiert ist. Das von Williams et al. (1984) im Modell EPIC erstmals beschriebene Konzept stellt ein Nettomineralisationsmodell dar. Die
Annahme ist, dass 80% der Streu leicht mineralisierbar ist und ohne Zwischenschritt zu
Nitrat umgebaut wird. 20% der Streu werden dem organischen N-Speicher zugeschlagen.
Für jeden Tagesschritt wird der Anteil an aktivem und stabilem organischem N berechnet, wobei in der Modellkonzeption lediglich N vom aktiven organischen N-Speicher zu
Nitrat mineralisiert werden kann. Wie aus der Abb. 1.2 ersichtlich, kann Ammonium in
diesem Modell nur durch die Zugabe von ammoniumhaltigem Dünger in den Boden gelangen, nicht jedoch als Zwischenprodukt der Mineralisierung entstehen. Weiterhin wird
in dieser Modellkonzeption postuliert, dass die Pflanzenwurzeln ausschließlich Nitrat aufnehmen und bei entsprechender Bodenfeuchte Denitrifikation stattfindet. Ein Teil des verbleibenden Nitrats steht dann für Transportprozesse zur Verfügung (Oberflächenabfluss,
Zwischenabfluss und Perkolation).
Eine kurze Beschreibung zum hydrologischen Modellkonzept von SWAT-G kann der
Einleitung aus Pohlert et al. (2005b) entnommen werden, welche im Kapitel 2, S. 27 zu
finden ist. Umfassende Modelldokumentationen zu SWAT und SWAT-G sind bei Neitsch
4
1.3 Aufbau des Einzugsgebietsmodells
et al. (2002) und Eckhardt et al. (2002) gegeben.
Für das Einzugsgebiets der Dill (AEO = 692 km2 ) wurde ein Datensatz aufgebaut,
der einerseits zur Bildung des Einzugsgebietsmodells und andererseits für die Kalibrierung/Validierung sowie des Post-Audits des Modells SWAT-G benötigt wird.
1.3.1 Das Dill Einzugsgebiet
Das Einzugsgebiet der Dill ist Teil des Rheinischen Schiefergebirges und befindet sich in
Mittelhessen. Abbildung 1.3 zeigt eine Übersicht über den Gewässerverlauf und die untersuchten Teileinzugsgebiete der Dill. Die Geologie im Süden des Einzugsgebiets besteht
aus karbonischem Tonschiefer und Grauwacke. Der mittlere Teil wird dominiert durch
devonisches Vulkangestein und Grauwacke. Im Norden herrschen devonische Sandsteine
aus Quarzit vor. Das Gestein des Einzugsgebiets ist an den Hängen bedeckt durch flachgründige, periglazial geprägte Böden, die durch den Prozess der Solifluktion im Pleistozän
gebildet wurden. Die Solifluktion führte zu einer erheblichen Anisotropie der hydraulischen
Leitfähigkeit der Böden, da der durch die Frostverwitterung herausgearbeitete Grobboden sich entlang der Hänge eingeregelt hat. Das heißt, die Längsachsen des Frostschutts
liegen parallel zur Hangoberfläche und bilden dadurch eine dachschindelförmige Struktur
im Untergrund aus (Fröhlich et al., 2005). Dies führt dazu, dass der Boden in hangparalleler Richtung höhere hydraulische Leitfähigkeiten aufweist als in vertikaler Richtung.
Diese Eigenschaft der periglazial geformten Böden wird im Modell SWAT-G durch den
Anisotropiefaktor ausgedrückt (Eckhardt et al., 2002). Der dominierende Bodensubtyp
ist die Typische Braunerde, und das anstehende Gestein ist durch geringe hydraulische
Leitfähigkeiten charakterisiert. In den Flusstälern haben sich Kolluvien entwickelt (Sauer,
2002).
Die durchschnittliche Lufttemperatur im Gebiet beträgt 9,5 C und die mittlere Jahresniederschlagsmenge 856 mm (1.7.2000 bis 31.12.2004). Das Verhältnis aus Abflussspende
und Jahresniederschlag für diese Periode ist k = 0, 48.
Detaillierte Angaben zur Problematik der Landnutzungsänderung in diesem Raum sind
bei Frede & Bach (1999) zu finden. Weitere Informationen zum Einzugsgebiet können der
Tabelle 1.1 entnommen werden.
1.3.2 Geobasisdaten
Als essentielle Geobasisdaten werden für die Einzugsgebietsmodellierung mit SWAT-G
Angaben zur Topographie, zu Bodeneigenschaften sowie zur Landnutzung benötigt. Aus
dem Datenpool des SFB299 standen ein digitales Höhenmodell (DHM) mit einer 25 X 25
m Auflösung (Hessisches Landesamt für Bodenmanagement und Geoinformation, HVBG),
eine Bodenkarte (BFD50, Hessisches Landesamt für Umwelt und Geologie, 2000) sowie
eine aus Fernerkundungsdaten (LANDSAT TM5) abgeleitete Landnutzungskarte für die
Jahre 1994/95 zur Verfügung (Nöhles, 2000).
5
a)
Dietzhölze
Obere Dill
1
3
2
Aar
4
Dill
0
Units (m) 10000
River gauges
5
1 Dietzhölze 3 Obere Dill 5 Aßlar (Outlet)
2 Dillenburg
4 Aar
Abbildung 1.3: Das Einzugsgebiet der Dill (Pohlert et al., 2006b, siehe auch Kapitel 4, S. 65).
Tabelle 1.1: Allgemeine Charakteristika des Einzugsgebiets der Dill (Pohlert et al., 2005b, siehe
auch Kapitel 2, S. 27).
Parameter
Topography
Value
Area
Maximum elevation
Minimum elevation
Mean slope
Maximum slope
Parameter
Land use
692 km2
675 m.a.s.l.
155 m.a.s.l.
8
35.2
Agriculture
Pasture
Changing vegetation
Forest, evergreen
Forest, deciduous
Urban area
Water
Climate
Annual precipitation
Annual discharge
Annual temperature
6
856 mm
457 mm
9.5 C
Value
7%
21%
9%
30%
25%
8%
< 1%
Demography
Population
148,000
Die BFD50 besteht aus vektorisierten Bodenkarten (1:50000), deren Geometriedaten
entsprechend der Blattschnitte als Kacheln vorliegen. Die kleinste Legendeneinheit ist eine Bodenform, die nach der deutschen Bodensystematik unterhalb einer Varietät steht
und nach Petrographie und Bodennutzung differenziert ist. Für jede Bodenform existiert
in der mitgelieferten Datenbank eine eindeutige Zuordnung zu einem Profildatensatz. Die
hydrologisch relevanten Parameter dieser Profildatensätze sind Lagerungsdichte des Bodens, die in fünf Klassen vorliegen, sowie die Bodenart, die als nominale Größe vorliegt.
Aus diesen beiden Angaben wurden mittels der Tabellen aus AG Boden (1994) die hydrologisch relevanten Bodeneigenschaften für jeden Horizont/Bodenschicht einer Bodenform
im Gebiet ermittelt (Friedrich & Lenhart, ohne Jahr).
Mittels der Software AVSWAT (Di Luzio et al., 2002) wurden aus dem DHM eine
Hangneigungskarte, ein topologisch konsistentes Gerinnenetz sowie eine Karte der Teileinzugsgebiete generiert. Um die HRU innerhalb der aus dem DHM generierten Teileinzugsgebiete zu bilden, wurden mit AVSWAT die Landnutzungskarte und die aufbereitete
Bodenkarte verschnitten. In der Konzeption von SWAT-G stellt ein HRU eine einheitliche
Kombination aus Bodeneinheit und Landnutzungstyp innerhalb eines Teileinzugsgebiets
dar, welches nicht zwangsläufig eine physisch zusammenhängende Fläche darstellen muss.
Die gleiche Kombination kann in einem anderen Teileinzugsgebiet wiederum existieren
und wird im Modell separat simuliert. Weiterführende Informationen zur GIS-gestützten
Aufbereitung für SWAT-G können dem Handbuch entnommen werden (Di Luzio et al.,
2001). Die aus den beschriebenen Geobasisdaten gewonnenen Eingangsdaten stellen die
Randbedingungen des Flussgebietsmodells dar.
1.3.3 Modellinput
Für den Aufbau eines Flusseinzugsgebietsmodells mit SWAT-G werden zumindest Tageswerte des Niederschlags sowie der Höchst- und Tiefsttemperatur der Luft benötigt. Da
in dieser Arbeit das Penman-Monteith Verfahren zur Berechnung der potentiellen Evapotranspiration angewandt wurde, wurden weiterhin Tageswerte der Globalstrahlung, der
mittleren Windgeschwindigkeit sowie der mittleren Luftfeuchte benötigt. Sämtliche meteorologischen Antriebsgrößen wurden auf Plausibilität nach dem von Meek & Hatfield
(1994) vorgeschlagenem Verfahren geprüft.
Weitere wichtige Antriebsgrößen für den Stoffhaushalt stellen die Ablaufwerte von
Kläranlagen, die N-Deposition aus der Atmosphäre sowie der Input von N durch
Düngemittel dar. Es wurden Monatswerte des Ablaufs von drei kommunalen Kläranlagen
sowie einem industriellem Einleiter erworben. Die insgesamt vier Punkteinleiter machen
etwa 90% des gesamten aus Punktquellen stammenden N im Einzugsgebiet aus (Lenhart,
2003). Eine konstante Nitratkonzentration im Niederschlag von 1,5 mg N L −1 wurden an
Hand des mittleren Jahresniederschlags und den Angaben der mittleren N-Deposition aus
den Karten von Gauger et al. (2001) ermittelt. Die Angaben zur tatsächlichen Landbewirtschaftung wie Fruchtfolgen und Düngegaben wurden aus Lenhart (2003) entnommen.
Zur Regionalisierung der an den Klimastationen gemessenen meteorologischen Größen
sieht AVSWAT das Nearest-Neighbour-Verfahren vor. Hierbei werden die Messwerte derjenigen Klimastation, die sich am nächsten zum Zentrum eines Teileinzugsgebiets befindet,
als räumlich repräsentativ für dieses Gebiet erachtet. Zur Höhenkorrektur der täglichen
7
Niederschlagssumme und der Tagestemperatur der Luft innerhalb eines Teileinzugsgebiets
wurden zehn Höhenschichten definiert, wobei ein Niederschlagsgradient von 3,54 mm d −1
km−1 und ein Temperaturgradient von -6,8 K d−1 km−1 für das Einzugsgebiet der Dill
ermittelt wurden. Die Gebietsniederschläge, bzw. die Gebietstemperaturen der Teileinzugsgebiete werden im Modell entsprechend den Höhenschichten und den angegebenen
Gradienten korrigiert. Da die meteorologischen Reihen zum Teil Messlücken aufwiesen,
bzw. der Betrieb einzelner Stationen während der Untersuchungsperiode (2000-2004) eingestellt wurde, wird die Aufbereitung der Stationswerte in den folgenden Unterabschnitten
gesondert beschrieben.
Niederschlag
Für das Einzugsgebietsmodell der Dill konnten aus dem Niederschlagsmessnetz des Deutschen Wetterdienstes (DWD) Tagessummen des Niederschlags von zwölf Niederschlagsstationen für die Periode 2000 bis 2004 erworben werden. Da während der Untersuchungsperiode drei der Niederschlagsstationen ihren Betrieb einstellten, wurden diese Reihen
mittels multipler linearer Regression synthetisch verlängert. Hierbei wurden die Werte
der Nachbarstationen als Prädiktoren verwendet. Die Kreuzvalidierung ergab einen RMSE zwischen 1,6 und 2,2 mm (Pohlert et al., 2006b, siehe auch: Kapitel 4, S. 64).
Temperatur
Für das Einzugsgebiet standen Werte der Tageshöchst- und -tiefsttemperatur der Klimastationen Aßlar, Dillenburg und Gießen zur Verfügung (2000-2004), wobei die letztgenannte Station außerhalb des Einzugsgebiets liegt. Messlücken der Temperaturreihe an
der Station Aßlar wurden mittels einfacher linearer Regression der Temperaturwerte der
Station Gießen geschlossen (RMSE = 1,13 C für Tageshöchsttemperatur und RMSE =
0,77 C für Tagestiefstemperatur).
Globalstrahlung / Sonnenscheindauer
Für diese Arbeit konnten Werte der Sonnenscheindauer für die Klimastationen Dillenburg
und Gießen erworben werden (2000-2004). An der Station Aßlar finden keine Messungen
der Sonnenscheindauer statt. Mittels der von Angstrøm (1924) gegebenen Beziehung zwischen dem Verhältnis aus Globalstrahlung (RS ) und extraterrestrischer Strahlung (RA )
sowie relativer Sonnenscheindauer (n/N ) wurden für die Stationen Dillenburg und Aßlar
tägliche Globalstrahlungswerte berechnet (Gl. 1.1). Die empirischen Koeffizienten a und
b wurden aus DVWK (1996) entnommen.
RS /RA = a + b × [n/N ]
(1.1)
Windgeschwindigkeit
Für die Stationen Dillenburg und Gießen lagen Tageswerte der mittleren Windgeschwindigkeit in der Einheit Bft (Beaufort) vor. Diese wurden mit der in DVWK (1996) an-
8
gegebenen Beziehung in die Einheit ms−1 umgerechnet und mittels des logarithmischen
Windprofils auf eine Referenzhöhe von 1,70 m gebracht.
1.3.4 Vergleichsgrößen
Für den Vergleich simulierter mit beobachteter Größen (Abfluss, N-Fracht) im Einzugsgebiet wurde sowohl auf vorhandene Beobachtungen des Hessischen Landesamtes für Umwelt
und Geologie (HLUG) als auch auf eigene Messungen zurückgegriffen.
Hydrographen
Im Einzugsgebiet der Dill befinden sich insgesamt fünf Pegel, die vom HLUG betrieben werden. Für die Untersuchungsperiode (2000-2004) lagen mittlere Tageswerte des
Abflusses vor. Die Pegel im Einzugsgebiet erfassen kontinuierlich den Wasserstand, welcher pegelspezifisch mittels empirisch gefundener Wasserstands-Abflussbeziehungen vom
HLUG in stündliche Abflusswerte umgerechnet werden. Aus den Stundenwerten werden
Tagesmittelwerte produziert. Der Pegel Aßlar stellt den Auslass des gesamten Gebiets
dar. Die Lage der Pegel und die entwässerten Teileinzugsgebiete sind den Abb. 1.3 sowie
4.2 im Kapitel 4 zu finden.
Nitratmessungen
Am Pegel Aßlar wurde im April 2000 ein automatischer Probenehmer (ISCO 3700, ISCO,
Lincoln NE, USA), der stündlich Gewässerproben aus der Dill entnommen hatte, von Lenhart (2003) installiert und bis zum Abschluss dieser vorliegenden Arbeit weitergeführt. Die
entnommenen Teilproben wurden volumengleich zu Tagesmischproben zusammengeführt,
durch 0,45 µm Polypropylen-Membranen gefiltert (Whatman puredisc, Whatman Inc.,
Clifton NJ, USA) und bei -20 C bis zur Analyse gelagert.
In der Zeit zwischen April 2000 bis April 2002 wurden die Gewässerproben photometrisch analysiert (Technicon Autoanalyzer N, Technicon Industrial Systems, Tarrytown
NY, USA). Details zu diesem Analyseverfahren sind in Kapitel 4, Abschnitt 4.2.5 zu finden. Ab Mai 2002 wurden die Gewässerproben mittels Ionenchromatographie (Dionex
DX-120, Dionex Corp., Sunnyvale CA, USA) nach DIN EN ISO 10304-1 bestimmt. Um
die systematischen Unterschiede der beiden Analyseverfahren zu berücksichtigen, wurden
paarweise Bestimmungen der Gewässerproben durchgeführt (n = 65). Die Messreihen
wurden mittels einfacher linearer Regression homogenisiert, wobei sich die Referenzreihe
auf die photometrisch bestimmten Werte beziehen (y = 0, 91 + 0, 83x, R 2 = 0, 72).
Zur weiteren Überprüfung des Modells wurden aus der Online-Ausgabe des Hessischen
Gewässergüteberichts (www.hlug.de) Monatswerte der Nitratkonzentration für den Pegel
Obere Dill (vgl. Abb. 1.3) recherchiert (2000-2003, n = 37).
Am 10.9.2003, 22.1.2004, 13.5.2004 und 14.9.2004 wurden jeweils Längsprofilmessungen
entlang der Dill sowie deren Zuflüsse durchgeführt. Diese Messungen dienten zur
Überprüfung, ob das Modell für einen gegeben Tag räumlich konsistente Ergebnisse produzieren kann. Ein solcher Ansatz wurde bis dato nur von Wade et al. (2002) für das Modell
INCA versucht, die jedoch lediglich ein mittleres jährliches Nitratkonzentrationsprofil des
River Kennet, UK, zur Überprüfung herangezogen hatten. Die Lage der Messstellen ist
9
der Abb. 4.2, S. 65 zu entnehmen. Somit standen für die Validierung des Modells insgesamt zwei Zeitreihen für die zeitliche Validierung am Oberlauf und am Gebietsauslass,
sowie vier Längsprofilmessungen für die räumliche Validierung zur Verfügung.
1.4 Test und Ablehnung des Modells
Die Überprüfung des Modellansatzes von SWAT-G fand schrittweise statt. Zunächst wurde das Einzugsgebietsmodell der Dill mit dem automatischen Kalibrieralgorithmus SCEUA (Duan et al., 1992) dahingehend kalibriert, dass der Quadratefehler zwischen simuliertem und beobachtetem Abfluss am Gebietsauslass minimiert wurde. Eine Beschreibung der Strategie zur Kalibrierung des Einzugsgebietsmodells ist sowohl in Kapitel 2
als auch in Kapitel 4 zu finden. Als weiterer Schritt wurden diejenigen Parameter manuell kalibriert, die den N-Umsatz im Modell SWAT-G steuern. Es wurden der Koeffizient
für die Humusmineralisation (CM N = 0, 003), der Koeffizient für die Mineralisierung
der Streu (RSDCO = 0, 005) und der Perkolationskoeffizient für den Nitrattransport
(N P ERCO = 0.4) kalibriert (vgl. Kapitel 2, Tabelle 2.2). Diese Parameter wurden zur
Kalibrierung gewählt, da Lenhart et al. (2002) in einer früheren Studie diese als sensitiv
eingestuft hatten.
Neben der Überprüfung des Einzugsgebietsmodells wurde mit SWAT-G ein Lysimetermodell aufgebaut und dieses ebenfalls überprüft. Hierzu wurden Datensätze aus einem
Langzeitversuch eines Düngeexperimentes in Brandis verwendet (Haferkorn, 2000). Der
Lysimeter wurde mit SWAT-G als ein einziges HRU innerhalb eines Einzugsgebiets simuliert, das durch eine Kombination aus Bodenprofil und ackerbaulicher Nutzung charakterisiert war. Als Hangneigung wurde 0% angenommen und das Lysimetermodell wurde
ohne Gerinne konstruiert. Eine weiterführende Beschreibung des Lysimeterdatensatzes ist
im Kapitel 3 zu finden. Die mit SWAT-G simulierte Evapotranspiration, Perkolation und
Nitratauswaschung wurden mit den Ergebnissen des Lysimeterexperimentes verglichen.
1.4.1 Ergebnisse und Diskussion
Zunächst sollen die Ergebnisse für das Einzugsgebietsmodell der Dill vorgestellt werden.
Für die Validierungsperiode (Juli 2000 bis Januar 2003) erzielte SWAT-G eine Modelleffizienz hinsichtlich der Größe täglicher Abfluss am Gebietsauslass Aßlar nach Nash &
Sutcliffe (1970) von E = 0, 56. Eine Diskussion dieser Modelleffizienz ist in Kapitel 2 zu
finden. Abbildung 1.4 zeigt die simulierte und beobachtete Monatsfracht des Nitrats für
den Pegel Aßlar. Die Modelleffizienz für die monatliche Nitratfracht beträgt E = 0, 66
für die Kalibrierungsperiode (März 2000 bis März 2002) und E = 0, 77 für die Validierungsperiode. Der Abb. 1.4 ist zu entnehmen, dass SWAT-G den jahreszeitlichen Verlauf
der Nitratfracht gut nachbildet. Die Spitzenfracht wird in den Wintermonaten erreicht,
was durch die hohe Mobilität des Nitrats im Boden zu dieser Jahreszeit im Einzugsgebiet
der Dill erklärt werden kann. Zum einen wird in dieser Jahreszeit durch die Winterruhe
nur wenig Stickstoff durch die Vegetation aufgenommen, zum anderen ist der Winter im
Einzugsgebiet der Dill die humide Jahreszeit mit entsprechend hohen Abflüssen. Beides
führt zu einer erhöhten Nitratauswaschung und somit erhöhten Nitratfrachten. Die mit
10
10000
8000
6000
Validation
Calibration
2000
4000
Observed
Simulated
0
Mean nitrate load [kg N d−1]
1.4 Test und Ablehnung des Modells
Mar 2000
Sep 2000
Mar 2001
Sep 2001
Mar 2002
Sep 2002
Mar 2003
Time [months]
Abbildung 1.4: Simulierte und beobachtete monatliche Nitratfracht (kg N d−1 ) mit SWAT-G
für die Periode April 2000 bis Januar 2003 am Pegel Aßlar (Pohlert et al.,
2005b, siehe auch Kapitel 2, S. 30)
SWAT-G an der Dill gefundenen Modelleffizienzen bezüglich simulierter Nitratfracht am
Gebietsauslass entsprechen denen anderer Modellierungsarbeiten über mesoskalige Flusseinzugsgebiete (z.B.: Chaplot et al., 2004; Grizzetti et al., 2003; Santhi et al., 2001a). Eine
ausführliche Diskussion und Einordnung der in diesem Schritt der Arbeit gefundenen Ergebnisse ist in Kapitel 2 gegeben.
Zusätzlich zur Validierung des Modells mit Nitratfrachten am Gebietsauslass wurden
die Simulationsergebnisse eines typischen ackerbaulich genutzten HRU analysiert. Tabelle
1.2 zeigt die mittleren jährlichen N-Flüsse für dieses HRU. Insbesondere die hohe jährliche
Denitrifikation (135,5 kg N ha−1 ) ist auffällig. Nach einer simulierten Düngung von 80 kg
N ha−1 wurden innerhalb von zwei Modelltagen 26 kg N ha−1 denitrifiziert und innerhalb
von neun Modelltagen 50 kg N ha−1 durch die Pflanze (Wintergerste) aufgenommen.
Sowohl die von SWAT-G simulierten Mengen als auch die Kinetik sind unrealistisch hoch.
Zur weiteren Prüfung der Gültigkeit des Modells SWAT-G wurde das Lysimetermodell
verwendet und mit Beobachtungen der monatlichen Nitratauswaschung in Brandis (Haferkorn, 2000) verglichen. Abbildung 1.5 zeigt den Verlauf der simulierten und beobachteten
Nitratauswaschung des Lysimeters (Profiltiefe: 3 m) in Brandis. Das Modell SWAT-G
simulierte nahezu keine Nitratauswaschung, obwohl deutliche Nitratauswaschungen am
Lysimeter beobachtet wurden. Die Ursache für die geringe simulierte Nitratauswaschung
liegt an der unrealistisch hohen Denitrifikationsrate des Modells SWAT-G und der damit
verbundenen nahezu vollständigen Zehrung des mineralischen N-Speichers im Modell.
11
Tabelle 1.2: Mittlere Jahresflüsse von N für ein typisches, ackerbaulich genutztes HRU im
Einzugsgebietsmodell der Dill simuliert mit SWAT-G. (Pohlert et al., 2005b, siehe
auch Kapitel 2, S. 32)
Process
Fertilisation
Denitrification
Plant uptake
Lateral flow
Leaching
Surface runoff
Nitrate kg N ha−1
151.7
135.5
44.3
23.9
4.4
5.1
50
10
20
30
40
Observed
Simulated
0
Nitrogen leaching (kg N ha−1 month−1)
B
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
Abbildung 1.5: Mit SWAT-G simulierte (grau) und beobachtete (schwarz) Nitratauswaschung
S. 55)
12
1.5 Formulierung und Test eines neuen Modells
1.4.2 Schlussfolgerung
Die bisherige Konzeption des N-Haushaltes von SWAT-G führte zwar zu akzeptablen
Ergebnissen hinsichtlich der monatlichen Nitratfracht am Gebietsauslass, jedoch weisen
die Simulationsergebnisse bei genauerer Betrachtung der internen Konsistenz des Modells
erhebliche Unplausibilitäten auf.
Der implementierte Ansatz zur Abbildung des N-Haushaltes weist folgende Schwächen
auf: Es handelt sich um ein stark vereinfachtes Nettomineralisationsmodell, das die Ammonifikation und die Nitrifikation vernachlässigt. Ferner ist die Denitrifikation stark vereinfacht abgebildet. Im Modell SWAT-G wird die Denitrifikation nur bei einer Bodenfeuchte größer oder gleich 95% der Feldkapazität simuliert (Neitsch et al., 2002, S. 187), wobei
die Kinetik der Denitrifikation dann ausschließlich durch die Bodentemperatur modifiziert
wird. Das Resultat dieser Konzeption sind unrealistisch hohe Denitrifikationsverluste und
eine rasche, sowie vollständige Zehrung des mineralischen Stickstoffspeichers im Boden
bei mäßiger Bodenfeuchte.
In SWAT-G ist ein Pflanzenwachstumsmodell implementiert, das die reale
Nährstoffaufnahme der Pflanze innerhalb eines Tages aus dem Nährstoffbedarf für optimales Wachstum der Pflanze und dem Nährstoffangebot aus dem Bodenspeicher berechnet. Deckt der Bodenspeicher den Bedarf, so wird ein optimales Pflanzenwachstum
simuliert. Ist der Bedarf größer als das Angebot, wird Nährstoffmangel simuliert (Neitsch
et al., 2002, S. 284). Unter N-limitierten Wachstumsbedingungen ist es jedoch in SWAT-G
möglich, dass sich über die Zeit ein Nährstoffdefizit der Pflanze akkumuliert. Wird dann
eine Düngung simuliert, kann nach dem Modell SWAT-G dieses Defizit sehr rasch von der
Pflanze gedeckt werden, was die hohe Kinetik der Nährstoffaufnahme durch die Pflanzen
erklärt. Die Konzeption von SWAT-G ist somit nicht in der Lage den N-Haushalt adäquat
abzubilden, was zu einer Ablehnung des Modells führen muss.
Um eine verbesserte Prozessbeschreibung des N-Umsatzes zu erzielen, wurden Konzepte
zur Mineralisation aus dem Denitrification-Decomposition-Modell (Li et al., 1992, 2000;
Zhang et al., 2002, DNDC, ) und zur Denitrifikation aus dem Modell CropSyst (Stöckle
& Campbell, 1989; Stöckle et al., 2003) verwendet. Die Konzeption des N-Haushaltes im
neuen Modell, das als SWAT-N bezeichnet wird, ist als Schema der Abb. 1.6 zu entnehmen.
1.5.1 Das Modell SWAT-N
Das Modell SWAT-N ist ein Bruttomineralisationsmodell, das drei organische N-Speicher
unterscheidet. Entsprechend der Kinetik sind dies ein sehr leicht abbaubarer, ein leicht
abbaubarer und ein schwer abbaubarer organischer N-Speicher. Jeder N-Speicher ist durch
ein spezifisches C/N-Verhältnis charakterisiert. Ein Teil des durch die Mineralisation, die
als Kinetik erster Ordnung beschrieben wird, freigesetzten Ammoniums wird durch den
Aufbau mikriobieller Biomasse immobilisiert. Überschüssiges Ammonium dient dem Aufbau des mineralischen Ammoniumspeichers in der entsprechenden Bodenschicht. Reicht
13
Abbildung 1.6: Neuer Ansatz zur Beschreibung des N-Kreislaufs in SWAT-N (Pohlert et al.,
2006a, siehe auch Kapitel 3, S. 39)
das mineralisierte Ammonium für das Mikrobenwachstum nicht aus, so wird zur Deckung des Bedarfs das notwendige N aus dem Ammonium- und Nitratspeicher gedeckt.
Ein einfaches Populationsmodell dient zur Simulation der Mikrobenmasse und somit des
mikrobiellen N-Bedarfs. Durch Absterben der Mikroben wird der in den Mikroben inkorporierte Stickstoff in Ammonium zersetzt. Ein Teil des Ammoniums wird an Tonteilchen adsorbiert. Entsprechend des vom pH-Wert des Bodens abhängigen AmmoniumAmmoniakgleichgewichts und der Diffussionskonstanten des Ammoniaks wird die Volatilisation berechnet. Weiterhin werden nitrifikatorische N-Verluste, die durch die Nitrifikation1 entstehen, ebenfalls berechnet. Das Konzept zur Beschreibung der Bruttomineralisation wurde aus den Arbeiten von Li et al. (1992, 2000); Stange (2001); Zhang et al. (2002)
entnommen und die mathematische Systembeschreibung ist bei Pohlert et al. (2006a)
sowie im Kapitel 3 zu finden.
Zur Berechnung der Denitrifikation wurde der mechanistische Ansatz von Stöckle &
Campbell (1989) verwendet. Dieser Ansatz lieferte in einer Vergleichsstudie, in der sechs
verschiedene Denitrifikationsfunktionen untersucht wurden, die akkuratesten Ergebnisse
hinsichtlich der Jahressumme gasförmiger N-Verluste. Die Denitrifikationsfunktion weist
jedoch eine Überschätzung der Häufigkeiten von Tagen mit Denitrifikation innerhalb eines
Jahres auf (Marchetti et al., 1997). Die Denitrifikation wird bei Stöckle & Campbell
(1989) als Summe sämtlicher gasförmiger, denitrifikatorischen Spurengase aufgefasst und
mit einer Reaktionskinetik erster Ordnung beschrieben. Die mathematische Beschreibung
ist ebenfalls bei Pohlert et al. (2006a) sowie im Kapitel 3 gegeben.
Ferner wurde das Pflanzenwachstumsmodell in SWAT-N dahingegen erweitert, dass die
Pflanzen sowohl Nitrat, als auch Ammonium zur Deckung ihres Nährstoffbedarfs aufnehmen können. Die tägliche Aufnahme von mineralischem N wurde mit einer Obergenze im
1
Auf die explizite Berechnung des instabilen Zwischenprodukts Nitrit wird in SWAT-N verzichtet.
14
Modell versehen, um den Fehler der unlimitierten N-Aufnahme bei Düngung und nach
einer längeren N-Stressphase zu korrigieren.
Sowohl für die Berechnung der Bruttomineralisierung als auch für die Denitrifikationsberechnung gilt, dass ausgehend von jeweils optimalen Bodenfeuchte- und Bodentemperaturbedingungen die Reaktionskinetik entsprechend der tatsächlichen Bodenfeuchte und
-temperatur mittels Reduktionsfunktionen verringert wird. Bodenfeuchte und Bodentemperatur pro Horizont/Schicht werden weiterhin durch den bisherigen Ansatz in SWAT-G
berechnet. Als neuer wesentlicher Randparameter wird der pH-Wert einer Bodenschicht,
bzw. eines Bodenhorizonts benötigt, der aus den neueren Versionen des BDF50 (Hessisches
Landesamt für Umwelt und Geologie, 2000) entnommen werden kann.
1.5.2 Sensitivitäts- und Unsicherheitsanalyse
Unter einer Sensitivitätsanalyse wird die Untersuchung einzelner Modellparameter oder
ganzer Modellparametersätze hinsichtlich ihrer Wirkung auf eine simulierte Ausgabegröße
oder mehrere Ausgabegrößen verstanden. Hierbei werden die Ausgabegrößen zunächst
nicht mit Beobachtungen des Systems verglichen. Die Sensitivitätsanalyse dient den Zwecken der Plausibilitätsprüfung und der Identifikation wichtiger steuernder Größen des
Modells, die in der Folge entweder genauer gemessen oder für eine Modellkalibrierung
herangezogen werden können.
Die Unsicherheitsanalyse im engeren Sinne umfasst die Untersuchung der Wirkung von
einzelnen Modellparametern oder Parametersätzen, die sich nicht oder nur schwerlich messen lassen, hinsichtlich ihrer Wirkung auf die Modellausgabe (Parameterunsicherheit).
Die Modellausgaben können hierbei mit dem Antwortverhalten des Systems verglichen
werden, wie dies u.a. beim Verfahren des Generalized-Likelihood-Uncertainty-Estimation
(GLUE, Beven & Freer, 2001) der Fall ist. Im weiteren Sinne wird unter der Unsicherheitsanalyse die Untersuchung der gesamten oder globalen Modellunsicherheit verstanden.
Hierunter fallen u.a. die numerische Unsicherheit, die Unsicherheit der Prozessabbbildung,
Parameterunsicherheit und die Messunsicherheit der beobachteten Referenzgröße des zu
simulierenden Systems. Um die globale Modellunsicherheit erfassen zu können, bieten
sich Ensemble-Modellierungen an, wobei unterschiedliche Computermodelle (Model Codes, vgl. Abb. 1.1, S. 2) das gleiche System zu simulieren haben. In dieser Arbeit richtet
sich der Fokus auf die Sensitivitätsanalyse und die Analyse der Parameterunsicherheit
von SWAT-N.
Zunächst wurde an Hand des Lysimetermodells mit SWAT-N eine regressionsbasierte Sensitivitätsanalyse durchgeführt, die erlaubt, Wechselwirkungen zwischen den Parametern zu berücksichtigen. Hierbei wurden zwölf Parameter ausgewählt, die im Modell
SWAT-N den N-Haushalt kontrollieren. Diese Parameter wurden jeweils aus Gleichverteilungen mittels randomisierter Monte-Carlo-Simulation (n = 4000) gezogen und hinsichtlich ihrer Wirkung auf die fünf Ausgabegrößen: Jahressumme der N-Ausgasung, Jahressume der N-Auswaschung, jährliche N-Aufnahme durch Pflanzen, mittlerer täglicher organischer N-Gehalt und mineralischer N-Gehalt des Bodens, untersucht. Für jede dieser
Zielgrößen wurden nach dem Backward-Selection Verfahren multiple lineare Regressionsmodelle erstellt, sowohl absolute als auch standardisierte Regressionskoeffizienten berechnet und anschließend der Größe nach aufgelistet. Im Kapitel 3, Abschnitt 3.5 sind die
15
Tabelle 1.3: Rangplätze der sensitiven Parameter für die jährliche N-Emission (NE ) und
jährliche N-Auswaschung (NL ). Es sind R2 das Bestimmtheismaß, bSD der standardisierte Regressionskoeffizient und bO der unstandardisierte Regressionskoeffizient
(Pohlert et al., 2006a, siehe auch Kapitel 3, Tabelle 3.5).
Variable
NE
R2
0.72
NL
0.78
Parameter
pH
k1
β
kDN
ERB
ENO
β
pH
k1
ERB
kDN
ENO
bSD
0.66
0.47
−0.27
0.07
0.07
−0.04
0.67
−0.52
−0.26
−0.07
−0.05
−0.02
Rangplatz
(1)
(2)
(3)
(4.5)
(4.5)
(6)
(1)
(2)
(3)
(4)
(5)
(6)
bO
11.3
14288.4
−20.5
373.9
19.5
−41.9
44.8
−7.7
−6801.0
−17.7
−205.4
−16.2
Rangplatz
(6)
(1)
(4)
(2)
(5)
(3)
(3)
(6)
(1)
(4)
(2)
(5)
Vorgehensweise, die ausgewählten Parameter, deren Bedeutung sowie die gewählten Parameterspannen dargestellt.
An dieser Stelle sollen auszugsweise die wichtigsten Ergebnisse der regressionsbasierten
Sensitivitätsanalyse für das Lysimetermodell vorgestellt werden. Eine umfassende Darstellung der Ergebnisse ist in Kapitel 3, Abschnitt 3.5 gegeben. Tabelle 1.3 zeigt die standardisierten und unstandardistierten Regressionskoeffizienten derjenigen Parameter, die
mittels der Regressionsanalyse als signifikante Prädiktoren für die jährliche N-Emission
und jährliche N-Auswaschung gefunden wurden. Für die jährliche N-Emission in die Atmosphäre sind der pH-Wert des Bodens (bSD = 0, 66), der Ratenkoeffizient für die Mineralisierung der sehr leicht abbaubaren Substanz (k1, b SD = 0, 47) und der Porenanteil
mit Anionenaustausch (beta) die drei sensitivsten Parameter, wobei der Parameter beta
negativ mit der Zielgröße korreliert ist (bSD = −0, 27). Für die jährliche N-Auswaschung
ist der beta-Parameter der sensitivste, gefolgt von den Parametern pH-Wert und k1 (Tabelle 1.3). Es soll an dieser Stelle festgehalten werden, dass für die Sensitivitätsanalyse
eine Parameterspanne für den pH-Wert von 3,5 bis 7,5 gewählt wurde.
Während in SWAT-N der Parameter k1 den absoluten Fluss von mineralisierbarem N
steuert, regelt beta die Menge des auswaschbaren Nitrats in den Bodenhorizonten. Dem
pH-Wert kommt eine Doppelfunktion zu: zum einen reguliert er neben der Bodentemperatur und -feuchte die Rate der Nitrifikation, zum anderen bestimmt er das Ammonium/Ammoniak Verhältnis und somit die mögliche Volatilisation. Im sauren Milieu wird
weniger und im neutralen Milieu wird mehr Ammoniak gebildet, weshalb der pH-Wert
positiv mit der jährlichen N-Emission korreliert.
Ferner wurde mit dem GLUE-Verfahren eine Unsicherheitsanalyse durchgeführt, deren Konzeption ausführlich in Kapitel 3, Abschnitt 3.6 erläutert ist. Die Strategie der
Monte-Carlo basierten Ziehung (n = 50000) der 12 Parameter war die Gleiche wie bei
der Sensitivitätsanalyse, jedoch wurde der pH-Wert aus einer Spanne zwischen 5,5 und
16
50
10
20
30
40
Observed
Upper bound
Lower bound
0
A
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
Abbildung 1.7: Mit SWAT-N simulierte (grau) und beobachtete (schwarz) Nitratauswaschung
S. 55)
6,5 gezogen, was eine typische Spanne für ackerbaulich genutzte Lößböden ist. Als Zielgröße wurde die monatliche Nitratauswaschung und als Zielfunktion die Modelleffizienz
E (Nash & Sutcliffe, 1970) gewählt. Entsprechend dem GLUE-Verfahren wurden die Simulationsläufe in akzeptable Realisierungen und inakzeptable Realisierungen unterteilt,
wobei E ≥ 0, 35 als Akzeptanzschwelle gewählt wurde. Zum Schluss wurden konditionelle
95-%-Vertrauensintervalle berechnet. Im Kapitel 3, Abschnitt 3.6 werden zusätzlich die
a-posteriori Verteilungen der Parameter erörtert.
Abbildung 1.7 zeigt wie die Abb. 1.5, S. 12 die simulierte und beobachtete monatliche
Nitratauswaschung des Lysimeters in Brandis. Es ist zu erkennen, das die Neuformulierung
der Konzeption des N-Haushaltes zu einer verbesserten Modelleffizienz geführt hat. Mit
dem neuen SWAT-N wurden Modelleffizienzen von bis zu E = 0, 4 gefunden, während der
alte Ansatz in SWAT-G nur zu einer Güte von E = −0, 22 führte.
1.5.3 Validierung von SWAT-N
Nach der Modellentwicklung und der Sensitivitäts- und Unsicherheitsanalyse an Hand
des Lysimeterdatensatzes wurde das neue Modell SWAT-N für das Einzugsgebiet der Dill
getestet. Hierzu wurden zunächst mittels Einzielkalibrierung (SCE-UA-Algorithmus) diejenigen Parameter kalibriert, die den Wasserhaushalt bestimmen. Für die Kalibrierung
wurden die täglichen Abflusswerte des Pegels Aßlar vom 1.7.2000 bis 31.1.2003 verwendet. Im Kapitel 4 wird ausführlich auf die Strategie bei der Kalibrierung eingegangen. In
der Tabelle 1.4 sind die kalibrierten Parameter und in der Tabelle 1.5 sind die erzielten
17
Tabelle 1.4: Kalibrierte Parameter des Einzugsgebietsmodells der Dill (1.7.2000 - 31.1.2003).
Die soil-unit 2458 stellt eine flachgründige Braunerde und Soil-unit 202 einen Auenboden dar. Die räumlich verteilten Parameter wurden nach Eckhardt & Arnold
(2001) kalibriert (siehe auch Kapitel 4, S. 69).
Parameter
Baseflow recession constant (d−1 )
Groundwater delay time
for aquifer reacharge (d)
Bulk density moist,
soil-unit 2458, layer 4 (g cm−3 )
Saturated hydraulic conductivity,
soil-unit 2458, layer 3 (mm h−1 )
Anisotropy, soil-unit 2458, layer 3
Lower limit
0.03
1
Upper limit
0.06
20
Calibrated value
0.034
19.8
2.51
2.64
2.64
10
85
83.3
1
45
44.9
2
8
4.65
Tabelle 1.5: Modelleffizienzen für simulierten und beobachteten täglichen Abfluss an verschiedenen Pegeln des Dilleinzugsgebiets (Kalibrierung: 1.7.2000 - 31.1.2003, Validierung: 1.2.2003 - 31.12.2004). E bedeutet Modelleffizienz nach Nash & Sutcliffe (1970), die Indices d, log, w, und m stehen für tägliche, log-Tageswerte,
wöchentliche und monatliche Werte (siehe auch Kapitel 4, S. 70).
Calibration
Validation
Ed
Elog
Ew
Em
Ed
Elog
Ew
Em
Aßlar
0.72
0.85
0.93
0.96
0.63
0.80
0.86
0.87
Dillenburg
0.76
0.84
0.92
0.97
0.74
0.74
0.89
0.92
Dietzhölze
0.66
0.71
0.89
0.91
0.56
0.60
0.81
0.84
Aar
0.66
0.14
0.79
0.80
0.68
-0.43
0.88
0.90
Obere Dill
0.51
0.14
0.67
0.74
0.66
0.36
0.87
0.94
Modelleffizienzen E für die Kalibrierung und Validierung (1.2.2003 bis 31.12.2004) aufgelistet. Eine Erläuterung, welche Modellfehler durch die Effizienzmaße gemessen werden,
ist in Kapitel 4 zu finden.
Für die Kalibrierungsperiode sind die Modelleffizienzen für die Pegel am Mittellauf (Dillenburg) und am Gebietsauslass (Aßlar) höher, als für Pegel in den Oberläufen (Dietzhölze,
Obere Dill, Aar). Wie zu erwarten, steigen die Modelleffizienzen bei sämtlichen Pegeln für
wöchentliche und monatliche Abflusswerte an, was durch den Glättungseffekt bei der Mittelung der Abflusswerte erklärt werden kann. Die Modelleffizienzen der logarithmierten
Abflusswerte zeigen, dass Abflüsse zu Niedrigwasserphasen für die Pegel Aßlar und Dillenburg sowohl für die Kalibrierungs-, als auch für die Validierungsphase akkurat simuliert
wurden. Im Gegensatz dazu ist die Modellgüte bei Niedriegwasser am Pegel Aar in der
Validierungsphase gering (Elog = −0, 43). Eine mögliche Ursache dafür ist, dass in dieser
Arbeit die Grundwasserparameter Baseflow Recession Constant und Groundwater Delay
18
Tabelle 1.6: Modelleffizienzen für die Nitratfracht (1.7.2000-31.12.2004) an den Stationen Aßlar
und Obere Dill. E bedeutet Modelleffizienz nach Nash & Sutcliffe (1970), D ist
der Index of Agreement (Wilmott et al., 1985), N̄o ist die mittlere beobachtete
Nitratfracht (kg N d−1 ), N̄s ist die mittlere simulierte Nitratfracht (kg N d−1 ),
sdo ist die Standardabweichung der beobachteten Werte (kg N d−1 ), sds ist die
Standardabweichung der simulierten Werte (kg N d−1 ) und n gibt die Anzahl der
Datenpaare wieder (vgl. auch Kapitel 4, S. 74).
Gauge
Aßlar
Obere Dill
Values
daily
log daily
weekly
monthly
daily
log daily
E
-0.29
0.25
0.22
0.65
-3.44
0.07
D
0.62
0.73
0.81
0.89
0.64
0.87
N̄o
2247
7.4
2231
2397
48
1.4
N̄s
2527
7.6
2523
2502
85
1.2
sdo
2649
0.8
2171
1896
54
0.6
sds
3003
0.6
2509
1690
141
1.0
n
1249
1249
223
55
37
37
Time lediglich als aggregierte Parameter für das gesamte Einzugsgebiet kalibriert wurden.
Da die Geologie in den Oberläufen Obere Dill und Aar in ihren Flächenanteilen deutlich
vom Gesamtanteil des Dilleinzugsgebiets abweicht, führt vermutlich die flächenaggregierte
Kalibrierung dieser Parameter für den Gesamtgebietsauslass zu einer Verschätzung in den
Oberläufen.
In der Tabelle 1.6 sind die Modelleffizienzen für die Simulation der Nitratfracht aufgelistet. Es soll an dieser Stelle nochmals betont werden, dass die Parameter, die den
N-Haushalt in SWAT-N steuern, für das Einzugsgebietsmodell nicht mehr kalibriert wurden. Das neue Modell liefert für die Station Aßlar gute Simulationswerte bei wöchentlicher
und monatlicher Mittelung, sowie nach der Logarithmierung der Tageswerte (Tabelle 1.6).
Letzteres bedeutet, dass die niedriegen bis mittleren täglichen Nitratfrachten von SWAT-N
gut reproduziert wurden. Aus diesen Ergebnissen kann geschlossen werden, dass SWAT-N
sowohl für den Oberlauf (Obere Dill), als auch für den Gebietsauslass (Aßlar) den Jahresgang der Nitratfracht richtig abbildet und räumlich konsistente Ergebnisse liefert. Für
Tageswerte liefert SWAT-N sowohl am Oberlauf (Obere Dill) als auch für den Gebietsauslass (Aßlar) unzufriedenstellende Ergebnisse. Im Kapitel 4, Abschnitt 4.3.2 sind weitere
Abbildungen zur visuellen Überprüfung der Modelleffizienz dargestellt.
Neben der Überprüfung simulierter Nitratfrachten mit zeitlich hochaufgelösten Messdaten wurden weiterhin simulierte Nitratkonzentrationen mit den Ergebnissen der zu
vier Terminen erfassten Längsprofilmessungen verglichen. Wie Abb. 1.8 zu entnehmen
ist, simulierte SWAT-N für die Abflusssituationen Hochwasser (22.1.2004, R 2 = 0.57),
Mittelwasser (13.5.2004, R2 = 0.30) und Niedrigwasser zum 15.9.2004 (R2 = 0.66) die
jeweiligen Flusslängsprofile der Nitratkonzentrationen mit akzeptabler Genauigkeit. Bei
Niedrigwasser macht sich insbesondere die Punkteinleitung des industriellen Einleiters
bemerkbar, was durch die Konzentrationsspitze an der Messstelle 28 deutlich wird. Für
die Niedrigwassersituation des 10.9.2003 kann SWAT-N das Längsprofil nicht nachbilden.
19
28
34
38
43
50
53
56
28
34
38
43
50
53
56
39
20
10
27
33
35
36
39
13 May 2004, intermediate flow
13 May 2004
34
38
43
50
53
56
10
57
27
33
35
36
39
15 Sep 2004, low flow
15 Sep 2004
28
34
38
Dill
43
50
53
56
57
51
0
10
20
Tributaries
NO3 (mg N l−1)
Dill
20
18
0
20
28
51
20
Tributaries
NO3 (mg N l−1)
Dill
18
51
0
57
10
12
36
22 Jan 2004
0
7
35
22 Jan 2004, high flow
10
12
33
Tributaries
0
7
20
27
NO3 (mg N l−1)
18
10
57
Dill
20
12
0
NO3 (mg N l−1)
20
10
18
10
7
NO3 (mg N l−1)
12
0
NO3 (mg N l−1)
7
NO3 (mg N l−1)
10 Sep 2003
0
NO3 (mg N l−1)
27
33
35
36
39
51
Tributaries
Abbildung 1.8: Vergleich zwischen simulierten (rot) und beobachteten (weiß) Nitratkonzentrationen diverser Messstellen entlang der Dill (links) und deren Zuflüsse
(rechts)(vgl. Kapitel 4, S. 77). Die Lage der Messstellen ist Abb. 4.2, S. 65
zu entnehmen.
Es soll an dieser Stelle betont werden, dass die Probenahme am 10.9.2003 gegen Ende
einer extremen Dürreperiode durchgeführt wurde. Die schlecht simulierten Nitratkonzentrationen sind für diesen Modelltag auf den fehlerhaft simulierten Abfluss zurückzuführen.
Dadurch ist im Einzugsgebietsmodell ein höheres Potential zur Verdünnung der Punkteinleitungen vorhanden, als dies in der Natur der Fall war. Mit einem Zeitversatz von 10
Tagen simulierte SWAT-N ebenfalls Konzentrationen in der Größenordung wie diese am
10.9.2003 beobachtet wurden (Daten nicht gezeigt). Eine ausführliche Diskussion der Ergebnisse, deren Einordnung mit Literaturangaben sowie die Sensitivität des Modells bei
Frost/Tauwetter auf die Mineralisation ist im Kapitel 4 gegeben.
20
1.6 Anwendung des neuen Modells
Tabelle 1.7:
Landnutzungsverteilung (%) des Dilleinzugsgebiets für die Referenz, das
AGENDA2000- und das CAP-Szenario (Pohlert et al., 2005a, S. 9).
Land use
Pasture
Lake
Urban
Forest
Arable land
Fallow land
Reference
20.7
0.3
9.1
54.8
6.0
9.2
AGENDA2000
25.5
0.5
9.5
56.3
8.0
0.2
CAP
30.4
0.5
9.5
55.8
3.7
0.2
1.5.4 Schlussfolgerung
Die Vergleiche zwischen den Modellausgaben des Lysimetermodells und den Lysimeterdaten einerseits, sowie des Einzugsgebietsmodells und den Datensätzen aus der Messkampagne im Dilleinzugsgebiet andererseits, lassen den Schluss zu, dass die Neuformulierung
des N-Umsatzes zu einer Verbesserung der Modelleffizienz und der internen Konsistenz
geführt haben. Die erzielten Modelleffizienzen lassen den Schluss zu, das SWAT-N ohne Kalibrierung der N-bestimmenden Parameter für die Anwendung und Bewertung von
Landnutzungsszenarien hinsichtlich monatlicher Nitratfrachten geeignet ist.
1.6 Anwendung des neuen Modells
Das neu entwickelte Modell SWAT-N wurde im Rahmen des integrierten Modellverbunds
des SFB299 für die Bewertung der Folgen der EU-Agrapolitiken AGENDA2000-Reform
und der neuen Common Agricultural Policy (CAP) für die Nitratfracht der Dill verwendet.
Bei CAP wird entgegen der AGENDA2000 Politik eine Entkoppelung der Agrarsubvention
von der erzielten Agrarproduktion favorisiert. Zunächst berechnete das agrarökonomische
Modell ProLand (Weinmann et al., 2005) unter der Annahme der Maximierung der
Flächenrente für die gegebenen agrarpolitischen Rahmenbedingungen jeweils eine vektorisierte Landnutzungskarte für das AGENDA2000- und das CAP-Szenario. Ferner wurden flächenspezifischen Angaben über die Bewirtschaftungssysteme als Attributwerte der
Flächenobjekte in den Karten geliefert.
Tabelle 1.7 zeigt die nach Nöhles (2000) als Referenz bezeichnete Landnutzungsverteilung des Dilleinzugsgebiets und die vom Modell ProLand prognostizierten Verteilungen
für das AGENDA2000- und CAP-Szenario. Es ist auffällig, das beim CAP-Szenario die
Grünlandnutzung gegenüber dem Referenzzustand und dem AGENDA2000-Szenario zunimmt, während die ackerbaulich genutzte Fläche abnimmt.
ProLand simuliert für jedes Flächenobjekt in der Landnutzungskarte eine individuelle Fruchtfolge mit Düngegaben. Um aus den ProLand-Ergebnissen Eingangsdaten der Bewirtschaftung für das Modell SWAT-N zu erstellen, wurde nach dem
Flächendominanzprinzip vorgegangen. Da ProLand für das AGENDA2000-Szenario auf
85% der Ackerfläche im Gebiet die Fruchtfolge Silagemais / Silagemais / Winterweizen
prognostizierte, wurde diese Rotation für SWAT-N auf sämtlichen Ackerflächen simu-
21
1984
1986
1988
1990
1992
1994
1996
1998
−3
−1
1
3
Difference between AGENDA2000 and reference scenario
−1
1
3
Difference between CAP and reference scenario
1984
1986
1988
1990
1992
1994
1996
1998
−3
Aerial nitrate load (kg N ha−1 mon−1)
Aerial nitrate load (kg N ha−1 mon−1)
Abbildung 1.9: Absolute monatliche Differenz der spezifischen Nitrafracht für das AGENDA2000 und CAP Szenario (Pohlert et al., 2005a, S. 10).
liert. Um Düngemengen zu erhalten, wurden Mediane der N-Gaben aus den Angaben der
Attributtabellen der Flächenobjekte ermittelt. Der Median der N-Gabe für die oben aufgeführten Kulturarten im AGENDA2000-Szenario beträgt 22 kg N ha−1 , 22 kg N ha−1 ,
132 kg N ha−1 . Das Grünland wurde als extensives Grünland mit einer N-Gabe von 33
kg N ha−1 parametrisiert. Für das CAP-Szenario wurde eine Fruchtfolge von Silagemais
/ Zuckerrübe / Winterweizen / Wintergerste für 85% des Ackerlandes von ProLand prognostiziert. Die Median der N-Gaben der einzelnen Kulturarten betragen 17 kg N ha −1 ,
148 kg N ha−1 , 103 kg N ha−1 , 32 kg N ha−1 . Grünland im CAP-Szenario ist extensiv
genutzt mit einer N-Gabe von 29 kg N ha−1 . Die Bewirtschaftung für das Refernzszenario
ist die Gleiche wie zur Validierung des Einzugsgebietsmodells (vgl. Kapitel 4). Für jedes
Szenario wurden für die Periode 1980 bis 2000 mit einer Vorlaufphase von 2 Modelljahren
monatliche Nitratfrachten simuliert.
Die mit SWAT-N simulierte mittlere spezifische Nitratjahresfracht beträgt 12,7 kg N
ha−1 für die Referenz, 13,4 kg N ha−1 für das AGENDA2000 und 11,6 kg N ha−1 für
das CAP-Szenario. Obwohl im CAP-Szenario die höchsten N-Düngegaben für Ackerland
22
1.7 Fazit und Ausblick
ausgewiesen wurden, liefert diese potentielle Landnutzung die geringste spezifische Nitratfracht. Dies ist auf den sehr geringen ackerbaulichen Flächenanteil (3,7%, vgl. Tabelle 1.7) zurückzuführen. Abbildung 1.9 zeigt die mit SWAT-N simulierten monatlichen
spezifischen Nitratfrachten als absolute Differenz zum Referenzszenario. Die monatlichen
Abweichungen zwischen den einzelnen Szenarien ist nur gering, was auf die extensive
Landnutzung des Dilleinzugsgebiets bei sämtlichen Szenarien zurückzuführen ist. Abbildung 1.9 ist zu entnehmen, dass die extremsten Unterschiede für einzelne Monate mit
bis zu 2 kg N ha−1 für das AGENDA2000-Szenario und bis zu -3 kg N ha−1 für das
CAP-Szenario zu verzeichnen sind.
1.7 Fazit und Ausblick
Zielsetzung dieser Arbeit war, ein szenarienfähiges Landschaftswasser- und
-stoffhaushaltsmodell zu entwickeln, das ohne Kalibrierung für die Prognose von
Nitratfrachten und -konzentrationen in mesoskaligen Einzugsgebieten angewandt werden
kann. Das Ergebnis dieser Arbeit ist ein weiterentwickeltes Modell SWAT-N, das sehr
detailliert die N-Mineralisation beschreibt und sowohl mit Lysimeterdaten, als auch mit
eigenständig erhobenen Messwerten im Einzugsgebiet der Dill geprüft wurde. SWAT-N
benötigt weiterhin eine Kalibrierung der abflussbestimmenden Parameter, was jedoch
durch die hohe Zahl von installierten Pegeln in Deutschland keine praktische Limitierung
darstellt.
Im Vergleich zum alten Ansatz in SWAT-G konnte sowohl die Richtigkeit, als auch
die Genauigkeit des Modells gesteigert werden. SWAT-N liefert akkurate Simulationsergebnisse für die monatliche Nitratauswaschung am Lysimeter sowie für monatliche und
wöchentliche Nitratfrachten am Gebietsauslass. Ferner bildet es Nitratkonzentrationsprofile entlang des Gewässers mit akzeptabler Genauigkeit ab. Auf Grund dieser Ergebnisse
wurde das Modell für szenarienfähig eingestuft und zur Bewertung von EU-Agrarpolitiken
hinsichtlich der Nitratfrachten im Rahmen des SFB299 herangezogen. Mit SWAT-N steht
nicht nur ein Forschungsmodell, sondern auch ein für die Maßnahmenplanung im Rahmen der Umsetzung der EU-WRRL praxisnahes Modell zur Verfügung. Sämtliche Eingangsgrößen des Modells werden in den EU-Mitgliedsstaaten routinemäßig erhoben (d. h.:
Pegeldaten, Klimadaten, DHM, Landnutzungskarten, Bodenflächeninformationssysteme,
Agrarstatistiken zur Konstruktion von Bewirtschaftungssystemen) und liegen somit für
die Modellierung vor.
Im Laufe der Bearbeitung stellten sich weitere Fragen, die in dieser Arbeit nicht beantwortet werden konnten. Sie sollen an dieser Stelle aufgeführt werden. Der Grundwasserzustrom und somit das über den Grundwasserpfad eingetragene Nitrat in SWAT-N
basiert auf einfachen Konzepten der Ganglinienseparation. Da im Einzugsgebiet der Dill
flachgründige Böden über schlecht wasserleitendem Gestein weit verbreitet sind, ist der
Grundwasserzustrom und somit grundwasserbürtige Anteil des Nitrats vernachlässigbar.
In anderen Flusseinzugsgebieten, in denen mächtige Porengrundwasserleiter vorhanden
sind, kann dieser Transportpfad einen deutlichen Einfluss auf die Chemie des Vorfluters
haben. Die Arbeiten von Sophocleous & Perkins (2000); Conan et al. (2003), in denen
das Modell SWAT mit dem Grundwasserströmungsmodell MODFLOW gekoppelt wur-
23
de, zeigen einen vielversprechenden Ansatz für die Problemlösung in Einzugsgebieten mit
entsprechenden Grundwassereigenschaften.
Ein wesentliches Phänomen trat bei der Simulation von Frühjahrsperioden mit SWAT-N
hinsichtlich der Überschätzung der Nitratfracht auf, was durch die rasche Mineralisierung abgestorbener mikrobieller Biomasse während Frost-Tau-Zyklen im Modell
zurückzuführen ist (vgl. Kapitel 4, Abschnitt 4.3.2). Der kontinuierliche Zusammenhang
zwischen Bodentemperatur und Mineralisationsraten im Wertebereich zwischen 5 C und
35 C ist in der Literatur gut beschrieben (z.B.: Leirós et al., 1999; Rodrigo et al., 1997;
Niklińska et al., 1999). Erst in jüngerer Zeit richtet sich die Aufmerksamkeit der Stickstoffforschung auf die episodisch an Frost-Tau-Zyklen gebundenen Spitzenemissionen von N 2 O,
bei denen innerhalb weniger Tage bis zu 70% der Jahressumme von N 2 O emittiert werden kann (Jungkunst et al., 2006). Die diskontinuierlichen Effekte von Frost-Tau-Zyklen
auf die Mineralisierung und darauf folgenden nitrifikatorischen und denitrifikatorischen
Verluste an die Atmosphäre sowie die Anreicherung von auswaschbarem Nitrat durch
diesen Prozess sind bis auf wenige Arbeiten (z.B. Ludwig et al., 2004; Müller et al.,
2002) noch unzureichend untersucht worden. Eine intensivere experimentelle Untersuchung dieses Phänomens mit dem Ziel der Prozessidentifizierung könnte zu einer weiteren
Verbesserung der Bruttomineralisationsmodelle und somit auch zur Verbesserung der Prognosefähigkeit von SWAT-N führen.
24
2 Pohlert et al., 2005, Adv. Geosci., 5,
7–12.
Der folgende Text wurde als Artikel in der Zeitschrift Advances in Geosciences im Jahre
2005 publiziert. Das Copywright liegt bei den Verfassern des Textes.
T. Pohlert, J. A. Huisman, L. Breuer, and H.-G. Frede, 2005. Modelling of
point and non-point source pollution of nitrate with SWAT in the river Dill,
Germany. Adv. Geosci. 5, 7–12.
Abstract. We used the Soil and Water Assessment Tool (SWAT) to simulate point
and non-point source pollution of nitrate in a mesoscale mountainous catchment.
The results show that the model efficiency for daily discharge is 0.81 for the calibration period (November 1990 to December 1993) and 0.56 for the validation period
(April 2000 to January 2003). The model efficiency for monthly nitrate load is 0.66
and 0.77 for the calibration period (April 2000 to March 2002) and validation period
(April 2002 to January 2003), respectively. However, the model efficiency for daily
loads is low (0.15), which cannot only be attributed to the quality of input data
of point source effluents. An analysis of the internal fluxes and cycles of nitrogen
pointed out considerable weaknesses in the models conceptualisation of the nitrogen
modules which will be improved in future research.
Keywords. nitrogen cycling; hydrologic modelling; model testing
2.1 Introduction
Eco-hydrological simulation models like the Soil and Water Assessment Tool (SWAT)
(Arnold et al., 1998) are essential tools for decision support in water resources planning
(e.g. to meet the requirements of the EU Water Framework Directive). We use SWAT
within the framework of the collaborative research center (CRC 299) ”Land use options
for peripheral regions”at the University of Giessen to assess the impact of potential land
use options on both the water quantity and quality.
Prior to the application of SWAT to predict the influence of land use change on river
water quality, the model is tested to observed data of discharge as well as nitrate load of
the river Dill. Both the accuracy and the reliability of the SWAT model with respect to
nitrogen cycling are discussed in detail in this paper.
25
2.2 Materials and Methods
2.2.1 The Dill catchment
The Dill catchment is located in mid-Hesse, Germany, and is part of the ”Lahn-DillBergland”(see Fig. 2.1). The geology of the catchment consists of carboniferous clay schists
and graywacke in the south, basic devonian volcanic rocks in the center and quarzitic
devonian sandstones and clay schists in the north. The dominant landform process during
the pleistocene was solifluction, thus, the rocks are widely covered by periglacial deposits.
Holocene colluviums appear mainly in river valleys. The dominant soil type is a shallow
cambisol (Sauer, 2002).
The Dill catchment faces a decline of farming, which causes an increase of fallow land.
Approximately 9 % of the catchment is fallow land, based on remotely sensed data in
1994 (Nöhles, 2000). Arable land contributes to about 7 % in the catchment. Table 2.1
summarises some of the main characteristics of the study site.
Figure 2.1: Stream network and locations of gauged subbasins within the Dill catchment.
26
2.2 Materials and Methods
Table 2.1: Characteristics of the Dill catchment. Information on topography is derived from the
DEM, land use distribution is taken from processed LANDSAT TM5 scenes (Nöhles,
2000) and climate data are taken from the hydrologic yearbook.
Parameter
Topography
Value
Area
Maximum elevation
Minimum elevation
Mean slope
Maximum slope
Parameter
Land use
692 km2
675 m.a.s.l.
155 m.a.s.l.
8
35.2
Agriculture
Pasture
Changing vegetation
Forest, evergreen
Forest, deciduous
Urban area
Water
Climate
Annual precipitation
Annual discharge
Annual temperature
960 mm
457 mm
8.3 C
Value
7%
21%
9%
30%
25%
8%
< 1%
Demography
Population
148,000
2.2.2 The SWAT model
SWAT is a semi-distributed, process oriented hydrological model. It is a continuous time
model which simulates both the water balance and the nutrient cycle with a daily time
step. The preprocessing of the spatially distributed data is achieved in a two-step approach. First the subbasins, stream network, channel length and hill slopes are derived
from a digital elevation model (DEM). Second, homogeneous land use and soil classes
are overlaid to yield multiple hydrological response units (HRU) within each subbasin.
Interpolation of weather data is achieved through the nearest-neighbour method, i.e. each
subbasin receives precipitation from one single rain gauge nearest to its centroid. Rainfall
within one subbasin is corrected for height with 10 elevation bands.
SWAT simulates evapotranspiration, infiltration, percolation, runoff generation, nutrient cycling and transport for each HRU. Water and sediment routing as well as in-stream
nutrient processes are simulated along the channel length for each subbasin (Neitsch et al.,
2002). The algorithms for nitrogen cycling and transport are based on the EPIC model
(Erosion-Productivity Impact Calculator) (Williams et al., 1984). Net mineralisation is
simulated with one active and one stable organic nitrogen pool. Plant uptake of nitrogen
is estimated using a supply and demand approach. Nitrate in the soil can be removed
from the soil via denitrification, mass flow of water as well as plant uptake (Neitsch et al.,
2002). In this study we used the SWAT-G model (Eckhardt et al., 2002), an adaption of
SWAT2000, which considers the anisotropy of vertical and lateral hydraulic conductivity
in mountainous regions.
2.2.3 Input data
The topography of the catchment was derived from a 25 x 25 m DEM. The land use map
was compiled from LANDSAT TM5 images taken in 1994 (Nöhles, 2000) and the soil map
is based on digitised soil maps with a scale of 1:50,000 (Hessisches Landesamt für Umwelt
und Geologie, 2000).
The meteorological input data were obtained from the German Weather Service (DWD).
27
Table 2.2: Manually calibrated parameters which control both nitrogen cycle and transport.
Parameter
CMN
RSDCO
NPERCO
Process
Humus mineralisation
Residue mineralisation
Nitrate transport
Initial
0.01
0.10
0.2
Final
0.003
0.005
0.4
The precipitation data consisted of homogenised daily records of 12 precipitation gauges.
Furthermore, daily records of solar radiation, minimum and maximum temperature, humidity as well as wind speed were taken from 2 climate stations. Data of monthly point
source effluents were taken from 3 municipal sewage treatment plants and one steel mill,
where nitric acid is used to harden steel. These four point sources comprise about 90 %
of total nitrate effluents from point sources in the catchment (Lenhart, 2003).
A three years crop rotation was simulated on arable land, where winter rape, winter
barley and oats were fertilised with 200, 120 and 80 kg N ha−1 , respectively. Pasture was
fertilised in the beginning of March and June as well as in mid-July and mid-November
with a total of 160 kg N ha−1 (Lenhart, 2003). A constant nitrate concentration in precipitation of 1.5 mg N l−1 was assumed.
2.2.4 Discharge and nitrate load measurements
An automatic sampler (ISCO 3700) was installed in April 2000 at the river gauge Asslar
(R 3462100, H 5605350). Daily composite samples of hourly subsamples are taken and
analysed in the laboratory. Lenhart (2003) used a photometric method to analyse NO 3 N for the period April 2000 until April 2002. From May 2002 onwards, the samples
are analysed with an ionchromatograph (DIN EN ISO 10304-1, 1995). A simple linear
regression was used to account for the systematic difference between the two methods
and to homogenise the series of nitrate concentrations,
N O3
IC
= N O3
PH
(0.87 ± (0.07))
R2 = 0.72
(2.1)
where N O3 IC and N O3 P H are NO3 N concentrations measured with the ionchromatograph and the photometer, respectively. Daily nitrate load is obtained by multiplying
daily average discharge with daily average nitrate concentration.
2.2.5 Calibration
Since nitrate fluxes strongly depend on water fluxes, parameters controlling water balance were calibrated in a first step. We used the Shuffled Complex Evolution algorithm
developed at the University of Arizona (SCE-UA) for automatic calibration of discharge
in the period from November 1990 to December 1993. The SCE-UA algorithm has proven
to be a powerful tool for finding the global minimum in the parameter space defined by
the user (Duan et al., 1992). The optimisation procedure as well as the chosen parameters
are identical to those reported by Huisman et al. (2004).
In a second step, the period from April 2000 to March 2002 was used to manually
calibrate daily nitrate load. The initial and final values of the calibrated parameters are
28
40
80
120
Observed
Simulated
0
Discharge [m3 s−1]
160
2.3 Results and Discussion
Jul 2000
Jan 2001
Jul 2001
Jan 2002
Jul 2002
Jan 2003
Time [d]
Figure 2.2: Simulated and observed daily discharge [m3 s−1 ] for the validation period (April
2000 to January 2003) at the gauge Asslar. E = 0.56, b = 0.81, R 2 = 0.60.
given in Table 2.2. These parameters were chosen based on the sensitivity analysis of
Lenhart et al. (2002).
Figure 2.2 shows simulated and observed daily discharge at the river gauge Asslar for
the validation period. The model efficiency (E) after Nash & Sutcliffe (1970) for daily
discharge is 0.81 and 0.56 for the calibration and validation period, respectively. In general
the model fits the observed rise of discharge as well as the recession well. However, the
timing and the height of single flood peaks in January 2001 and January 2003 are less
accurately predicted. This can be attributed to the model structure of SWAT. Since it is a
continuous time model with a daily time step, subscale processes such as flood generation
can not be efficiently simulated. Furthermore, there is a temporal mismatch between daily
precipitation measured for 24 h beginning from 7:30 AM and daily discharge averaged over
24 h from midnight on.
Figure 2.3 gives simulated and observed monthly nitrate loads for both the calibration
and validation period. Table 2.2 shows the coefficient for humus mineralisation (CMN)
and the coefficient for residue mineralisation (RSDCO), which needed to be reduced by
one to two orders of magnitude from the default values to slow down the simulated
kinetics. For the calibration and validation period E is 0.66 and 0.77, respectively. The
peak load of nitrate occurs during wintertime. This can be attributed to both the high
runoff during this season, which leads to increased nitrate transport as well as the lack
of plant uptake, which favors the generation of leachable nitrate. SWAT simulates the
29
10000
8000
6000
Validation
Calibration
2000
4000
Observed
Simulated
0
Mean nitrate load [kg N d−1]
Mar 2000
Sep 2000
Mar 2001
Sep 2001
Mar 2002
Sep 2002
Mar 2003
Time [months]
Figure 2.3: Simulated and observed monthly nitrate load [kg N d−1 ] for the period April 2000
to January 2003 at the gauge Asslar. Calibration: E = 0.66, b = 1.26, R 2 = 0.69.
Validation: E = 0.70, b = 1.21, R2 = 0.77.
seasonal cycle accurately. However, SWAT did not match the temporal patterns of daily
loads, thus the model efficiency for daily load is low (E = 0.15).
The contribution of point source pollution during low flow phases in summer can account
for up to 90% of the total nitrate load at the outlet. In winter the contribution is lower
(Fig. 2.4). We assumed a simple mixing model to derive the ratio between the sum of
point source inlets along the channel and the total load at the river outlet. It should
be mentioned, that during summertime the ratio between total effluents and observed
nitrate load can exceed 100% which is indicated in Fig. 2.4 by missing data. This can be
attributed to the simple mixing model, which ignores degradation processes in the river,
as well as the considerable uncertainty of the aggregated point source inlet data.
The model performance regarding nitrate load prediction is in accordance with reports
from other authors using SWAT on various catchments. Lenhart et al. (2003) worked on
the same catchment and obtained a model efficiency for the validation period (April 2001
to March 2002) of 0.09 and 0.31 for daily and monthly loads, respectively. In contrast to
the present study, where point source effluents were used as input data for the simulation,
Lenhart et al. (2003) used annual estimated point source effluents based on a correlation
with population data. These loads were used to calculate a point source background
concentration by division with simulated daily runoff. This background concentration was
subtracted from the observed concentrations. The adjusted concentration was then used to
calculate daily loads for validation and calibration. Chaplot et al. (2004) predicted mean
monthly nitrate loads in the Walnut Creek watershed (51.3 km 2 ), Iowa. The coefficient of
determination (R2 ) in their work is 0.73. Santhi et al. (2001a) found a similar agreement
between simulated and observed monthly data (R 2 = 0.72, E = 0.64) when they used
30
100
20
40
60
80
Observed
Simulated
0
Contribution of point sources [%]
Mar 2000
Sep 2000
Mar 2001
Sep 2001
Mar 2002
Sep 2002
Mar 2003
Time [months]
Figure 2.4: Monthly contribution of point sources as a fraction of observed and simulated total
nitrate load at the gauge Asslar.
SWAT to predict nitrate loads in the Bosque River watershed (4277 km 2 ), Texas. However,
Grizzetti et al. (2003) found a lower agreement (E = 0.30) when they used SWAT at the
Vantaanjoki watershed (1680 km2 ), Finland, to model diffuse emissions and retentions of
nutrients on a monthly basis.
As outlined above, we achieved a similar simulation efficiency of nitrate load at the
catchment outlet as other authors working with SWAT. However, we examined the internal fluxes and cycles of nitrate in the catchment in addition. In Fig. 2.5 the rates of
denitrification and plant uptake are given for one HRU where winter barley grows on
agricultural land. Within the first two days after fertilisation of 80 kg N ha −1 about 26
kg N ha−1 are denitrified and about 50 kg N ha−1 nitrate are taken up by the plant in
the first nine days. Both the amount and kinetics are suprisingly high. The high kinetics
in plant uptake can be explained by the lack of a sink limitation for the simulated nitrate
uptake. If the nitrate content of the soil does not supply the nitrate demand of the crop,
a deficiency will occur and accumulate as long as this condition continues. As soon as the
nitrate pool in the soil is increased by fertilisation, the plant can take up as much as its
accumulated deficiency, causing the high kinetics.
Table 2.3 gives average annual values for the main nitrate fluxes in the same HRU as
used for Fig. 2.5. It is obvious that the amount of simulated denitrification is far too
high. In SWAT denitrification occurs when soil moisture exceeds 95% of field capacity.
Since the soils are not highly conductive and the water is often perched on the underlying
bedrock layer, this threshold is often exceeded. Hence, leaching and denitrification are
two competing processes within the model.
31
80
20
40
60
Fertilisation
Denitrification
Plant uptake
0
Nitrate [kg N d−1]
100
0
2
4
6
8
Days after fertilisation
Figure 2.5: Denitrification and plant nitrate uptake of winter barley after fertilisation with 80
kg N ha−1 .
Table 2.3: Average annual values for the main nitrate fluxes on an example HRU with agricultural land use.
Process
Fertilisation
Denitrification
Plant uptake
Lateral flow
Leaching
Surface runoff
32
Nitrate kg N ha−1
151.7
135.5
44.3
23.9
4.4
5.1
2.4 Conclusions
2.4 Conclusions
We used SWAT to simulate both point and non-point source nitrate pollution in a
mesoscale catchment. The model was able to accurately simulate the seasonal cycle of
nitrate load. The agreement between simulated and observed daily nitrate load was lower.
The poor accuracy of simulated daily loads can not only be explained by the limited availability of daily data of point source effluents. Since denitrification rates and daily plant
uptake are misspredicted, the conceptualisation of the internal nitrogen cycling seems to
be the main limitation. Clearly, this model set up can not yet be used for scenario analysis. Besides an accurate result for the outlet, a consistent representation of internal N
cycling is also required. Otherwise, scenario analysis might result in unrealistic predictions
of changes in N budgets due to land use change.
In our ongoing research, the algorithms of organic litter distribution between 3 different pools, gross mineralisation, ammonification, denitrification as well as plant uptake
considering a source and sink limitation from the Denitrification-Decomposition Model
(DNDC) (Li et al., 1992, 2000; Zhang et al., 2002) are implemented and tested in SWAT
to overcome the aforementioned problems.
2.5 Acknowledgements
The research project is granted by the German Research Foundation (DFG) in the framework of the Collaborative Research Center (CRC) 299 at the University of Giessen.
33
34
3 Pohlert et al., 2006, Ecol. Model., in
press.
Der folgende Text wurde bei der Zeitschrift Ecological Modelling am 22. Dezember 2005
eingereicht und am 17. November 2006 durch den Herausgeber zur Veröffentlichung
akzeptiert. Das Copywright liegt bei dem Verlag Elsevier.
NOTICE: this is the author’s version of a work that was accepted for publication in
Ecological Modelling. Changes resulting from the publishing process, such as peer review,
editing, corrections, structural formatting, and other quality control mechanisms may
not be reflected in this document. Changes may have been made to this work since it was
submitted for publication. A definitive version was subsequently published in Ecological
Modelling, DOI:10.1016/j.ecolmodel.2006.11.019.
T. Pohlert,J. A. Huisman, L. Breuer, H.-G. Frede, 2006. Integration of a
detailed biogeochemical model into SWAT for improved nitrogen predictions
— model development, sensitivity, and GLUE analysis. Ecol. Model., in press,
doi:10.1016/j.ecolmodel.2006.11.019.
Abstract. In this study the Soil and Water Assessment Tool (SWAT) was extended
with algorithms from a detailed nitrogen turnover model to enhance the model
performance with regard to the prediction of nitrogen leaching. The new model,
which is further referred to as SWAT-N, includes algorithms for decomposition,
growth of nitrifying bacteria, nitrification, nitrificatory as well as denitrificatory Nemissions, N-uptake by plants and N transport due to water fluxes. The model was
tested with a lysimeter dataset of a long term fertilisation experiment including crop
rotation conducted in Eastern Germany.
A regression based global sensitivity analysis was employed to test the impact of
the new implemented parameters on the sensitivity of various model output variables. The rate coefficient for decomposition, the pH-value, and the porous fraction
from which anions are excluded were identified as the most important parameters
controlling nitrogen leaching and gaseous nitrogen emissions.
A generalised likelihood uncertainty estimation (GLUE) was conducted afterwards
to calculate conditioned prediction intervals for each simulated time step. A maximum model efficiency after Nash & Sutcliffe (1970) of 0.4 could be achieved for the
simulation of monthly nitrogen leaching. It is concluded, that the implemented algorithms enhance the model performance of SWAT, since the previous SWAT version
failed to accurately simulate nitrogen leaching at the investigated site.
Keywords. N-Budget; Lysimeter; Denitrification; Mineralisation; Nitrification; Uncertainty estimation; Sensitivity analysis
35
3.1 Introduction
There is an increasing demand for an integrated management of catchment wide water
resources focusing both on water quantity and quality in various water bodies such as
streams, lakes and ground water throughout the world. In the European Union the Water
Framework Directive (2000/60/EG) was established in the year 2000, which claims an
integrative catchment management to achieve a “good ecological status” regarding the
nutrient status of water bodies. In principal, eco-hydrological models provide essential
tools for the management of river catchments, since simulations of potential water management strategies, climate change effects, as well as land use change effects on water
quality and water quantity can be made. Results of these simulations can be used to
support decision makers before action plans are implemented, provided that the chosen
model structure considers all relevant processes of the soil-vegetation-water system and
simulates them accurately.
There are various models available which simulate hydrological processes, nutrient
turnover including plant growth on the land phase, nutrient transport through surface
runoff, interflow, and groundwater flow, as well as further in-stream nutrient turnover on
the catchment scale. The conceptualisation of these models vary with respect to complexity, and, consequently, their demand for parameters and their computational efficiency.
They can be grouped depending on their spatial representation of processes as semidistributed or fully distributed models.
The Integrated Nitrogen model for Catchments (INCA, Whitehead et al., 1998a) is
an example for a semi-distributed non-point source pollution model. INCA focuses on
diffuse N-emissions from arable land, provided that an external water balance model
calculates the required hydrological input variables. A second semi-distributed model is
the HSPF model (Bicknell et al., 1997), which includes a conceptual water balance model
and simulates in-stream processes of nitrogen in particular. The HBV-N (Bergström, 1995;
Arheimer & Brandt, 1998) and LASCAM model (Viney et al., 2000) are developed for
large scale watersheds (A ≥ 10, 000 km2 ), where water and nutrient cycles are simulated
on the basis of subbasins. In contrast, the Soil and Water Integrated Model (SWIM,
Krysanova & Haberlandt, 2002), which is a spin-off of the Soil and Water Assessment
Tool (SWAT, Arnold et al., 1998), simulates runoff generation, nutrient turnover, plant
growth, and farm management on the basis of Hydrological Response Units (HRU).
Apart from the above mentioned semi-distributed models the MIKE-SHE model (Refsgaard & Storm, 1995) can be classified as a fully distributed hydrological model, which
simulates water cycle, water quality and sediment erosion. Other fully distributed models
for water quality prediction, are the TNT model of Beaujouan et al. (2001), which was
developed for the simulation of N-emissions from rural areas, and the recently published
Diffuse Nitrate Modelling Tool (DNMT, Liu et al., 2005), which includes the simulation of
N leachate from the sewerage in urban areas. These two models follow the TOPMODEL
approach to simulate hydrological fluxes. A 3D-model for the simulation of nitrogen transformation and transport based on the advection–dispersion equation is the SHETRAN
model (Birkinshaw & Ewen, 2000), which can be used for small scale catchments (A ≤ 10
km2 ). Constraints of the fully distributed physically based watershed models are both their
high demand for computational power and their high demand for data input. Hence, these
36
3.2 Model description
models can often only be employed in small and well investigated research watersheds.
Recent studies show that the accuracy for the simulation of non-point source pollution
on rivers can be improved through the coupling of detailed N transformation models with
semi-distributed hydrological models. Li et al. (2004) coupled DAYCENT with SWAT for
an improved simulation of long term mineral N export of forested catchment. Cui et al.
(2005) linked the Wetland-DNDC model with MIKE SHE to simulate C and N dynamics
with respect to greenhouse gas emissions of a pine tree forest in Florida. Borah & Bera
(2003) as well as Horn et al. (2004) provide comprehensive reviews of various existing
models for the simulation of catchment wide nonpoint-source pollution.
In this study, we focus on the widely used and accepted SWAT model. The SWAT
model was chosen in this study because it provided accurate predictions of daily discharge in mountainous catchments in Germany in previous studies (e.g. Huisman et al.,
2004). Several recent applications (e.g. Grizzetti et al., 2003; Bouraoui et al., 2004; Santhi
et al., 2001b; Lenhart et al., 2003) have shown the capability of SWAT to predict N export
at the catchment outlet. However, as pointed out in Pohlert et al. (2005c), the original
conceptualisation of N dynamics in SWAT has several limitations, which might lead to
an unrealistic representation of internal N-processes (e.g. high denitrification losses). Although a correct representation of N processes is not necessarily required for an adequate
prediction at the catchment outlet, it is generally accepted that this is a prerequisite when
models are used in scenario analysis. To overcome these problems with the SWAT model,
we extended SWAT with algorithms from the DNDC (Li et al., 1992, 2000) and the CropSyst model (Stöckle et al., 2003). The DNDC model is a well accepted biogeochemical
model with a detailed conceptualisation of decomposition. The relevant algorithms which
were included into SWAT are provided in the following sections.
In this study, we used the SWAT2000 version (Neitsch et al., 2002). SWAT is a semidistributed continuous time model for the simulation of both the water and nutrient cycle
of meso-scale river catchments. For the set-up of a site-specific model, the watershed is
delineated from a digital elevation model and further divided into subbasins. Unique combinations of land use and soil types within a subbasin are aggregated to form Hydrologic
Response Units (HRU) (Neitsch et al., 2002). The processes of evapotranspiration, infiltration, surface runoff, subsurface flow, percolation, sediment erosion, crop growth and
N cycle are simulated for each HRU. The resulting water, nutrient and sediment fluxes
from each HRU are accumulated within their corresponding subbasin and allocated to
the main reach of the subbasin. Discharge and matter fluxes are routed within the stream
network from one subbasin to another and finally to the outlet of the watershed using
either the variable storage routing method (Arnold et al., 1995) or the Muskingum river
routing method.
In this study the N cycle of the land phase (i.e. HRU) is examined. Figure 3.1 illustrates
the original approach implemented in SWAT, where N cycle is based on the EPIC model
(Williams et al., 1984). The conceptual approach of SWAT distinguishes one passive
and one active organic N pool. SWAT is a net-mineralisation model, because organic
37
Figure 3.1: Original conceptualisation of the N cycle in SWAT
N from the active pool is directly converted to nitrate, whereas nitrificatory gaseous N
losses are not modelled explicitly. These N losses are implicitly taken into account due
to the parametrisation of the rate constants for mineralisation. In this conceptualisation,
ammonium can only occur in a given soil layer when ammonium fertiliser is applied.
Plants can only take up nitrogen in the form of nitrate.
This concept shows several theoretical weaknesses. First, the net-mineralisation approach of SWAT does not take the intermediate N-compount ammonium into account.
Nitrificatory N losses that occur during the stepwise oxidation of organic N into ammonium, nitrite and nitrate are ignored in this concept. Within this concept, the processes of
mobilisation and immobilisation can not be simulated. Furthermore, in the SWAT concept
plants can only take up N in the form of nitrate, although it is known that ammonium
can also be a substantial N source for various crops.
In the new model, the N description in Fig. 3.1 is replaced by algorithms from the
Denitrification-Decomposition model (DNDC, Li et al., 1992, 2000) as well as the CropSyst
model (Stöckle et al., 2003). Figure 3.2 gives the new conceptualisation of the modelled
N cycle. The model distinguishes three pools of organic litter, two pools of microbes,
and two pools of humads. Each pool is characterised by specific C/N ratios and rate
coefficients. The model is a gross-mineralisation model, since a part of the mineralised N
is used for microbial growth (immobilisation) and nitrificatory N-emissions are considered
explicitly. Mineralisation is simulated in a stepwise fashion, where first ammonium is build
(ammonification) and after that, then nitrate is formed (nitrification). Gaseous emissions
38
Figure 3.2: New conceptualisation of the N cycle in SWAT-N.
of N2 O and NO are taken explicitly into account. The intermediate product nitrite is not
explicitly modelled in this concept. The ammonium is partly adsorbed on clay particles;
the remaining ammonium can be mineralised or taken up by plants. Denitrification is
calculated with a function given by Stöckle & Campbell (1989).
A detailed description of the process conceptualisation is given in the following sections;
the reader is referred to Tables 3.6 and 3.7 for a definition of parameters and their values.
The new model is further referred to as SWAT-N.
3.2.1 N input into the soil system
The input of nitrogen in the modelled soil system is considered due to wet atmospheric
deposition and fertilisation. In contrast to SWAT where only wet deposition of nitrate was
taken into account, the new SWAT-N model also considers the deposition of ammonium:
FNO3 ,l=1 = NP f(NO3 :NTOT )
FNH4 ,l=1 = NP 1 − f(NO3 :NTOT )
(3.1)
(3.2)
where FNO3 ,l=1 and FNH4 ,l=1 are NO3 and NH4 input in the corresponding pools of
the top soil layer (l = 1) (kg N ha−1 d−1 ), NP is the mineral nitrogen content of daily
precipitation (kg N ha−1 d−1 ) and f(NO3 :NTOT ) is the fraction of N which is nitrate in the
daily precipitation.
Furthermore, mineral N fertiliser will be added into the corresponding mineral N pools
39
of the top soil, if fertilisation takes place at a given day. Organic fertiliser is allocated into
the organic C pools depending on its C/N ratio as described in the following section.
3.2.2 Allocation of organic matter to organic litter pools
The model distinguishes three organic litter pools which are characterised by specific
C/N ratios and decomposability. The allocation of organic matter (i.e. plant residues and
organic fertiliser) to the three pools is calculated as suggested by Li et al. (2000). Table
3.7 provides the pool specific C/N ratios used for this study.
Since the allocation of organic litter depends on the C/N ratio of residues, Eq. 3.3 was
implemented in SWAT-N to calculate the carbon content of residues. It is assumed that
the mass of dry matter of a crop mainly comprises of carbon, nitrogen and phosphorous.
CRES = MRES − NRES − PRES
(3.3)
where CRES is the organic carbon in crop residues (kg C ha−1 ), MRES is the biomass of
residues as dry matter (kg ha−1 ), NRES is the nitrogen content of residues (kg N ha−1 ),
and PRES is the phosphorous content of residues (kg P ha−1 ).
3.2.3 Decomposition of soil organic matter
Decomposition is the production of mineral C and ammonium from soil organic matter
mediated by microorganisms. The algorithms for decomposition are taken from Li et al.
(1992, 2000). Each organic litter pool is characterised by both a specific C/N ratio and
a specific rate constant ki . The gross decomposed carbon for any litter pool is calculated
as given by Eq. 3.4:
FC,l = ki f(T,l θ,l) Ci,l
(3.4)
where FC,l is gross decomposed organic carbon available for growth of nitrifying bacteria
(kg C ha−1 d−1 ) at soil layer l, ki (i = 1, . . . , 3) are rate coefficients for i = 1 very fast,
i = 2 fast, and i = 3 slow decomposable organic matter at optimum temperature and soil
moisture, Ci,l is the organic carbon content of pool i in layer l (kg C ha−1 ), and f(T,l θ,l)
is the reduction factor accounting for suboptimal temperature and moisture conditions
(Eq. 3.5). The combined reduction factor is calculated in accordance to Stange (2001):
1
1
f(T,l θ,l) = 2 /
(3.5)
+
fθ,l fT,l
where fθ,l is the soil moisture factor and fT,l is the temperature factor for layer l,
respectively. Prior to the calculation of the soil moisture factor, the volumetric soil moisture content is converted to Water Filled Pore Space (WFPS, Eq. 3.6). After that, the
soil moisture factor is calculated with a Weibull-function (Stange, 2001):
Al =
40
θv,l
φl
(3.6)
fθ,l
−1
Al − MCRIT,l
= 1 − 1 + exp
γ
(3.7)
where Al is the water filled pore space in layer l, θv,l is the volumetric water content in
layer l (m3 m−3 ), φl is the porosity, MCRIT,l is the half optimum soil moisture (WFPS),
and γ is a scale parameter of the Weibull-function.
The temperature reduction function (Eq. 3.8) is calculated with the O’Neill-function
as proposed by Stange (2001):
fT,l =
T
TMX − TACT,l A
TMX − TOPT
TACT,l − TOPT
× exp TA
TMX − TOPT
(3.8)
where TACT,l is actual temperature of layer l, TMX is maximum temperature for decomposition, TOPT is optimum temperature for decomposition, and TA is a scale parameter
of the O’Neill-function. The actual soil temperature TACT,l is calculated according to the
original version of SWAT (Neitsch et al., 2002, p. 42–46).
The gross decomposed carbon FC,l is partly used for the formation of microbial biomass.
The amount of incorporated nitrogen can be calculated assuming a constant C/N ratio
of microbes. If the released ammonium due to decomposition meets the daily demand
for microbial growth, a net mineral surplus will occur which fills the ammonium pool.
Otherwise the microbes will reduce nitrate to meet their nitrogen demand, consequently,
a net mineral nitrogen loss will occur. The total net ammonification during a day F NH4 ,l
is the sum of four individual fluxes. Equations 3.9 to 3.12 describe the individual fluxes
of net decomposed organic nitrogen.
FC,l + DBC,l
− DBN,l
(3.9)
F1NH4 ,l =
CNB
Where DBC,l is the surplus of microbial biomass in layer l (kg C ha−1 d−1 ), CNB is the
C/N ratio of microbes, and DBN,l is the microbial N demand for biomass formation (kg
N ha−1 d−1 ).
In addition to the decomposition of organic matter decaying microbial biomass further
contributes to the formation of ammonium:
DRB,l (1 − EAC − ENO )
(3.10)
CNB
where F2NH4 ,l is the net flux of ammonium from decomposed microbes into the ammonium pool of layer l (kg N ha−1 d−1 ) , DRB,l refers to total decomposed microbes (kg C
ha−1 d−1 ), EAC is the efficiency coefficient for microbes decomposition, and E NO is the
efficiency coefficient for resistant humads formation.
The third relevant source for ammonification is the decomposition of resistant humads
as given by Eq. 3.11:
F2NH4 ,l =
F3NH4 ,l =
DHC,l 0.8 (1.0 − CLl )
CNH
(3.11)
41
where F3NH4 ,l is the net flux of ammonium from decomposed humads into the ammonium pool (kg N ha−1 d−1 ), DHC,l refers to decomposed humads (kg C ha−1 d−1 ), CLl is
the fraction of clay content of layer l, and CNH refers to the C/N ratio of humads.
During frost events a part of the microbe population will die and, consequently, contributes to the ammonium surplus.
F4NH4 ,l =
DDEAD,l
CNH
(3.12)
where DDEAD,l is the amount of dead microbes due to a frost event of layer l (kg C ha −1
d−1 ). For the calculation of DDEAD,l the reader is referred to Li et al. (2000).
3.2.4 Clay adsorption
A part of the released ammonium is adsorbed on clay minerals and, hence, is not involved
in further N turnover. The clay adsorption is calculated after Li et al. (1992):
ANH4 ,l = (0.41 − 0.47 log SNH4 ,l )
CLl
CLMX
(3.13)
where ANH4,l is adsorbed ammonium in layer l (kg N ha−1 d−1 ), SNH4 ,l is the ammonium
content of soil layer (kg N ha−1 ), CLl is clay content (%) and CLMX is the maximum clay
content. The remaining ammonium SNH4 ,l can be further nitrified, volatilised or taken up
by plants, as described in the following sections.
3.2.5 Ammonia volatilisation
Nitrogen in the form of ammonia has a very low partial pressure which leads to evaporation
of ammonia under common atmospheric conditions. The amount of ammonia in a specific
soil layer is given by the ammonium – ammonia equilibrium (Eq. 3.14):
cNH4 ,l
log KNH4 − log KH2 O = log
+ pHl
(3.14)
cNH3 ,l
where KNH4 is the constant for dissociation of ammonium, KH2 O is the constant for
dissociation of water, cNH4 ,l and cNH3 ,l are ammonium and ammonia concentration of
layer l (mg l−1 ), respectively, and pHl refers to the pH-value of layer l. The daily rate of
ammonia volatilisation is calculated after Li et al. (1992):
AM = 2 SNH3
p
DNH3 /3.14
(3.15)
where AM is volatilised ammonium (kg N ha−1 d−1 ), SNH3 is the actual ammonia
content (kg N ha−1 ), and DNH3 is the ammonia diffusion coefficient (cm soil ha kg−1 d−1 ).
3.2.6 Nitrification
Nitrification is a multi step oxidation process catalysed by several autotrophic organisms
such as Nitrosomas ssp. and Nitrobacter spp.. Zhang et al. (2002) proposed the following
first order kinetics to estimate gross nitrification:
42
FNO3 ,l = SNH4 ,l
h
i
−kNI fTNI,l )
(
1 − exp
fθNI,l fphNI,l
(3.16)
where FNO3 ,l is the gross nitrified ammonium (kg N ha−1 d−1 ), kNI is the rate coefficient
for nitrification, fTNI,l is the temperature reduction factor for nitrification (Eq. 3.17), f θNI,l
is the moisture reduction factor (Eq. 3.18), and fphNI,l is the pH-specific reduction factor
(Eq. 3.19). The temperature reduction factor is calculated according to:
fTNI,l =
T
TMXNI − TACT,l ANI
TMXNI − TOPNI
TACT,l − TOPNI
× exp TANI
TMXNI − TOPNI
(3.17)
where TMXNI is the maximum temperature of the O’Neill-function for nitrification, T OPNI
is the optimum temperature for nitrification, and TANI is a scale parameter. The moisture
reduction function for nitrification fθNI,l is calculated as a 4th order polynomial according
to Zhang et al. (2002), where Al is the soil moisture content (WFPS).
fθNI,l = −12.904 A4l + 17.651 A3l
− 5.5368 A2l + 0.9975 Al − 0.0243
(3.18)
The pH-specific reduction value fphNI,l is given by a 2nd order polynomial (Zhang et al.,
2002), where pHl is the pH-value for layer l.
fphNI,l = −0.0604 pHl2 + 0.7347 pHl − 1.2314
(3.19)
The emission of nitrous gases due to nitrification is given by Li et al. (2000) (Eqs. 3.20
and 3.21):
a =
(34.6 − 9615)
(TACT,l + 273.15)
NOX,l = kOXi FNO3 ,l θACT,l 2.72a
(3.20)
(3.21)
where NOX,l is either N2 O or NO production (kg N ha−1 d−1 ), kOXi is a coefficient for
either N2 O production or NO production.
3.2.7 Denitrification
Denitrification is the reduction of nitrate to gaseous N compounds namely N 2 O, NO
and N2 under anaerobic conditions. The original denitrification function implemented in
SWAT is based on EPIC (Williams et al., 1984), where the rate of denitrification is a
function of varying soil temperature and a constant organic C content of the soil layer.
Denitrification only occurs, when a specific threshold of moisture is exceeded. In the
43
current version of SWAT this threshold is set to 95 % of the water content at field capacity.
This leads to unrealistically high denitrification losses in cases of relatively moist soils,
where this threshold is often exceeded. Pohlert et al. (2005c) showed that N-leaching and
denitrification are two strongly competing processes within the current conceptualisation
of SWAT. Due to the cascaded percolation model, water only percolates into an underlying
soil layer, when field capacity is exceeded. Thus, denitrification occurs before water starts
flowing which leads to extraordinary high N losses (up to 135 kg N a −1 ) and a complete
depletion of nitrate in the pools of each soil layer.
To overcome the aforementioned problems with the SWAT conceptualisation, we implemented the mechanistic denitrification function from CropSyst model (Stöckle & Campbell, 1989; Stöckle et al., 2003). Denitrification is now modelled as a function of soil
moisture status and temperature. In a comparative study, where six denitrification functions were analysed, this function was found as the most reliable one, although it tends
to overestimate the frequency of denitrification occurrences and underestimates denitrification rates (Marchetti et al., 1997). Denitrification is calculated based on a first-order
kinetics ignoring the stepwise reduction of nitrate to other nitrogen compounds (Stöckle
& Campbell, 1989):
DN = SNO3 ,l [1 − exp (−kDN fθDN,l fTDN,l )]
(3.22)
where DN is the gaseous N loss due to denitrification during a day (kg N ha −1 d−1 ),
kDN is the coefficient of denitrification, fDN,θ is the dimensionless moisture factor and
fDN,T is the temperature factor. The moisture factor is calculated according to:
fθDN,l = exp 0.304 + 2.94 (θSAT,l − θACT,l ) − 47 (θSAT,l − θACT,l )2
(3.23)
where θSAT,l is the volumetric water content at saturation (m3 m−3 ) in layer l and θACT,l
is the actual volumetric water content (m3 m−3 ). The temperature factor is calculated for
soil temperatures below 10
as follows:
fTDN,l = 0.67 exp [0.43 (TACT,l − 10)]
and for soil temperatures above 10
∀
TACT,l ≤ 10
∀
TACT,l > 10
(3.24)
:
fTDN,l = exp [0.08 (TACT,l − 15)]
(3.25)
3.2.8 Plant uptake of N and N fixation
The conceptualisation of N-uptake of plants follows the original SWAT version. The optimal N content of a crop in SWAT is a function of its growth stage given optimal growth
conditions. The plant nitrogen demand for a given day is the difference between the optimum N content of the plant as a function of its phenology and the plants actual N content.
This difference will then be taken up from the rootable soil profile, where a depth distribution of nutrient uptake depending on root density is taken into account. The SWAT-N
model assumes that the crop prefers N in the form of nitrate, but takes up ammonium
as soon as the nitrate pool in a given soil layer is depleted. In situations were the actual
44
3.3 Data set for model testing
mineral N content of the soil does not meet the crop’s nitrogen demand, nitrogen stress
will occur.
Legumes are able to fixate N directly from the atmosphere. In SWAT a legume fixates N
in situations, where the mineral N content of the soil does not meet the crop’s N demand.
It is assumed, that the crop directly incorporates N into its biomass. The original concept
of SWAT was not changed in this study. The reader is referred to Neitsch et al. (2002)
for a detailed description of algorithms regarding N plant uptake and N fixation.
3.2.9 N transport
Apart from crop N-uptake and N removal due to harvest as well as gaseous N losses,
N can leave the soil system due to percolation, lateral flow as well as surface runoff. In
this study the original conceptualisation of nitrate movement in the soil was used. SWAT
distinguishes a mobile and an immobile fraction of nitrate in each soil layer (Neitsch et al.,
2002). The mobile nitrate can leave the top soil layer due to surface runoff, percolation,
and lateral flow. The mobile nitrate in the underlying soil layers can move by lateral flow
and percolation, only. The algorithms for calculating organic N losses from the top soil
by surface runoff are given by Neitsch et al. (2002).
3.3 Data set for model testing
After the modification of the model structure, a site specific model was created with a
data set taken from a long-term fertilisation experiment conducted at the research station
Brandis in Saxony, Germany (Haferkorn, 2000). To match the spatial representation of
process description in SWAT-N, the lysimeter was modelled as one single HRU with a
hill slope of 0 and without a stream network. Consequently, only the processes such as
N-turnover in the soil, N-uptake by crops and vertical movement of N within the soil
column was considered in the model (apart from additional in-stream processes in a full
catchment model). This site specific model is in agreement with the definition of an HRU
within SWAT, because the lysimeter is an unique combination of land use, soil type, and
crop management.
The data set consists of monthly percolation, actual evapotranspiration and mineral
N leachate for the period 1980 to 1992 of three replicates from a lysimeter experiment
to minimise the effect of single measurement errors. The adjacent fields of the lysimeters
received the same land management (i.e. cropping and fertilisation) to minimise a potential
overestimation of evapotranspiration due to the oasis effect. The soil columns were taken as
undisturbed soil samples and the soil can be classified as a Haplic Arenosol, which consists
of sandy loess above fluvial sand with gravel (Knappe et al., 2002). Furthermore, precise
information regarding management practice (i.e. crop rotation, sowing dates and harvest,
time and amount of fertilisation) were included as well as annual totals of atmospheric
wet N-deposition and N-uptake of crops. The dimensions of the lysimeter are 1 m 2 of
surface area and 3 m of depth. The large depth of the lysimeter ensures that the effect
of an artificially created discontinuity of suction at the interface between the soil column
and the bottom of the lysimeter vat becomes minimal. Hence, the construction of the
45
Table 3.1: Soil physical parameters of lysimeter group 5 (Haplic Arenosol) at Brandis
(after Haferkorn, 2000). z is depth of layer, CL is clay content, SL is silt content,
HU M means humus content, KSAT is saturated hydraulic conductivity, and BD is
wet bulk density.
Horizon
z
CL SL HU M
KSAT
BD
(cm)
(%) (%)
(%)
(mm hr−1 ) (g cm−3 )
Ap
0 – 35
8.0 30.0
2.1
635.0
1.49
C1
35 – 175
2.0
2.0
782.9
1.67
C2
175 – 300 4.0
2.0
391.7
1.53
lysimeter does not erroneously mask the measurement of percolation (Haferkorn, 2000).
Table 3.1 shows the soil physical details of lysimeter group 5. Haferkorn (2000), Knappe
et al. (2002) and Dreyhaupt (2000) provide further information about the experimental
design.
Daily records of humidity, wind velocity, maximum and minimum temperature were
obtained from the weather station of Leipzig (approx. 15 km northwest of Brandis). Precipitation and sunshine duration were measured directly at the field. Global radiation
(RG ) was calculated according to Angstrøm (1924), given the relation between R G and
extra terrestrial radiation as well as the proportion of bright sunshine for a specific location. Rainfall was corrected after Richter (1995) to account for the systematic wind error
which occurs when making measurements with standard rain gauges. During the study
period annual average precipitation and annual average actual evapotranspiration ranged
between 390 to 727 mm (mean 594 mm) and 370 to 488 mm (mean 436 mm), respectively;
annual N loss due to percolation ranged between 14 and 98 kg ha −1 N a−1 (mean 47 kg N
ha−1 a−1 ) and seasonal crop uptake of N varied between 50 to 155 kg N ha−1 a−1 (mean
103 kg N ha−1 a−1 ). Crop management and climatic totals are provided in Table 3.2.
3.4 Model calibration
Prior to the global sensitivity analysis (GSA), a multi-criteria calibration was performed
considering parameters which control evapotranspiration and percolation. A stepwise calibration is justified, since the simulated N percolation of a coupled nutrient cycle and
transport model strongly depends on the accuracy of simulated water flows. If no calibration of water flow is conducted, the sensitivity of parameters controlling N cycling will
likely be masked by the inappropriate representation of soil moisture and water fluxes.
The MOSCEM-UA algorithm (Vrugt et al., 2003) was employed for model calibration.
A classical split sampling test was conducted using the period from 1980 to 1986 for
model calibration and the period from 1987 to 1992 for model validation. Details of the
calibration strategy can be found in Huisman et al. (2005). Table 3.3 provides initial and
calibrated values of the considered parameters. The initial values were taken from the
Brandis dataset (see also Table 3.1). For actual evapotranspiration, the model efficiencies
(NSE, Nash & Sutcliffe, 1970) are 0.746, 0.715 and 0.732 for the calibration, validation
and total period, respectively. Figure 3.3 provides the observed and simulated time series.
For percolation, the NSE for the calibration, validation and entire period are 0.584, 0.632
46
3.5 Sensitivity analysis
Table 3.2: Crop rotation, precipitation (P ), actual evapotranspiration (ET ), N-fertilisation
Year
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
(NF ), N-deposition (ND ), N-leaching (NL ), and N-uptake by crops (NUP ) for the
lysimeter in Brandis (after Haferkorn, 2000; Dreyhaupt, 2000). Both P and E T refer to the period November of the preceding year to October of the specified year,
respectively.
Crop
P
ET
NF
ND
NL
NUP
(mm (mm (kg N
(kg N
(kg N
(kg N
a−1 ) a−1 ) ha−1 a−1 ) ha−1 a−1 ) ha−1 a−1 ) ha−1 a−1 )
Maize
657
140
44
97
Sugar beet
727
478
160
53
88
111
Winter wheat 390
395
120
28
60
90
Winter barley 672
488
120
33
47
117
Grass
536
452
175
42
14
118
Potato
477
381
100
69
17
73
Winter wheat 581
387
120
35
61
94
Potato
629
449
100
37
98
131
Winter wheat 574
370
140
46
56
72
Winter barley 546
469
120
46
34
124
Sugar beet
579
487
140
44
21
155
Winter wheat 417
427
140
37
25
101
Winter barley 583
445
120
29
35
50
and 0.609, respectively (Fig. 3.4).
So called One-Factor-at-a-Time sensitivity analyses were intensively used for SWAT as
well as DNDC by various authors (e.g. Lenhart et al., 2002; Stange, 2001). In this type
of sensitivity analysis, the sensitivity is measured by the response of one single output
variable to a fractional change of one single input parameter, while other parameters
remain constant. An inevitably constraint of this methodology is the implicit ignorance of
higher order effects such as compensation and synergism. Hence, a regression based global
sensitivity analysis based on Monte-Carlo simulation was employed. This methodology has
been successfully applied for various complex hydrological models (Christiaens & Feyen,
2002; Sieber & Uhlenbrook, 2005; Muleta & Nicklow, 2005).
3.5.1 Regression based global sensitivity analysis
The GSA was employed with a Monte-Carlo sampling strategy, where 12 parameters as
given in Table 3.6 were simultaneously sampled (N = 4, 000) from uniform distributions
except k1 , which was drawn from a log-uniform distribution. This was done, because the
uncertainty of k1 is large (see Table 3.6) and a log-uniform distribution for parameters that
have a large uncertainty range ensures a more efficient Monte-Carlo based sampling of
high dimensional parameter spaces (Beven & Freer, 2001). Parameters were selected since
47
Table 3.3: Initial and calibrated parameter values. Parameters of the top soil layer were cali-
120
100
80
60
40
20
0
Evapotranspiration (mm month−1)
brated and the parameters of underlying soil layers were adjusted according to the
change ratio of the top soil layer as proposed by Eckhardt & Arnold (2001).
Parameter Definition
Initial value Calibrated value
−3
BD
Bulk density (g cm )
1.490
1.479
AW C
Available water capacity (m3 m−3 )
0.168
0.188
α
Albedo
0.250
0.219
KSAT
Saturated hydraulic
conductivity (mm hr−1 )
635
747
ESCO
Soil evaporation
compensation factor
0.000
0.528
EP CO
Plant uptake
compensation factor
0.000
0.297
Calibration
|
|
|
|
|
|
|
|
|
|
|
|
|
Validation
Observed
Simulated
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
Figure 3.3: Observed and simulated monthly evapotranspiration at lysimeter station Brandis,
Germany (Data taken from Haferkorn, 2000).
48
|
|
|
|
|
|
|
|
|
|
|
|
|
20
40
60
80
100
Calibration
0
Percolation (mm month−1)
120
Validation
Observed
Simulated
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
Figure 3.4: Observed and simulated monthly percolation at lysimeter station Brandis, Germany
(Data taken from Haferkorn, 2000).
their sensitivity on model output was not focused in DNDC related literature hitherto.
For each of the 4,000 parameter sets, the modified SWAT-N was run. Both the MonteCarlo simulation and the multiple regression analysis were conducted with the software
package R, version 2.0 (R Development Core Team, 2004).
A stepwise multiple linear regression (backward selection) was applied using the parameter sets derived from the Monte-Carlo simulations as predictor variables and several
simulated components of the N cycle as the dependent variables. The following components of the N cycle were taken into account: annual average atmospheric N emission
(NE ), average annual N-leaching due to percolation (NL ), average annual nitrogen uptake
by crops (NUP ), mean daily organic nitrogen content (NORG ) and mean daily mineral
nitrogen content (NMIN ). Table 3.4 provides statistics for the simulated components of
the N cycle. It is worth to note that both simulated NL as well as simulated NUP are
within a similar range as compared to the observations of 47 kg N ha −1 a−1 and 103 kg N
ha−1 a−1 , respectively. Furthermore, NE that here denotes the sum of N2 , NOX and NH+
3
comes closer to an approximate value of 13 kg N ha−1 a−1 assuming a default emission
factor of 0.1 (kg NH3 -N + NOX -N emitted / kg N applied) for synthetic fertiliser that is
added to the soil as it is the case in the IPCC-method (IPCC, 1997).
The coefficient of determination R2 measures the proportion of explained variance of
the dependent variable due to the included predictors provided that a linear relation exists. The linearity was tested by analysing the distribution of the residuals, which form a
Gaussian distribution with a mean of zero in the case of a linear relation. The ordinary
regression coefficient (bO ) for each individual predictor is a direct measure of sensitivity
49
Table 3.4: Quantiles of analysed output variables. Annual gaseous nitrogen emission is N E ,
annual nitrogen leaching is NL , and nitrogen crop uptake is NUP , average daily
organic nitrogen content is NORG and mineral nitrogen content is NMIN (n = 4, 000).
Quantile
NE
NL
NUP
NORG
NMIN
(kg N
(kg N
(kg N
(kg N
(kg N
ha−1 a−1 ) ha−1 a−1 ) ha−1 a−1 ) ha−1 d−1 ) ha−1 d−1 )
0%
2.6
5.8
91.7
479.4
18.2
25%
20.4
17.4
102.0
529.2
66.5
50%
43.1
30.1
102.7
548.9
98.4
75%
61.6
45.8
103.9
571.2
168.7
100%
82.5
92.5
110.
657.7
353.5
and indicates the rate of change in units of the criteria, when the predictor is changed
by one unit while the other predictors remain constant (Christiaens & Feyen, 2002). Additionally, the standardised regression coefficient (bSD ), where both criteria and predictor
are standardised by their corresponding standard deviation, allows a direct comparison
of the relative importance of each parameter. The sign of both b SD and bO indicate the
direction of change of the criteria due to the change of the predictor variable.
3.5.2 Results of the sensitivity analysis
Table 3.5 provides the results of the regression based GSA. As indicated by R 2 , the
linear regression model explains a relatively high proportion of the total variance of the
parameters NE , NL , NORG and NMIN . However, only 53 % of the variance of NUP can be
explained by the linear regression model. Since the distribution of the residuals for this
specific multiple regression model is significantly different from a Gaussian distribution, it
is concluded that a strong non-linear effect exists on NUP . This can be attributed due to
the fact that a crop rotation was simulated with SWAT-N, where each crop is determined
by a unique set of crop parameters. Hence, the crop parameters and the weather variables
during a growth season seem to be more important for crop growth in the model and,
thus, simulated nitrogen uptake than the chosen parameters that mainly govern turnover
rates and transport of nitrogen.
The parameters kNI , kRB , khrb , DNH3 and DRF (for definition see Table 3.6) are found
insensitive, since their corresponding regression coefficients are not significantly different
from 0.
A strong positive impact of k1 on gaseous nitrogen emission is indicated by a high
regression coefficient (bSD ). The higher k1 the more organic nitrogen can be decomposed,
which leads to higher nitrification N losses as well as ammonia volatilisation.
50
Table 3.5: Ranking of sensitive parameters of various output variables. R 2 denotes coefficient
of determination, bSD is standardized regression coefficient, bO means ordinary regression coefficient and Rank means absolute numerical importance (N = 4, 000).
For the definition of other symbols refer to the text as well as Table 3.6.
Output
NE
R2
0.72
NL
0.78
NUP
0.53
NORG
0.78
NMIN
0.69
Parameter
pH
k1
β
kDN
ERB
ENO
β
pH
k1
ERB
kDN
ENO
pH
β
k1
ERB
kDN
k1
ENO
β
ERB
pH
EAC
β
pH
k1
ERB
kDN
bSD
0.66
0.47
−0.27
0.07
0.07
−0.04
0.67
−0.52
−0.26
−0.07
−0.05
−0.02
−0.52
−0.50
−0.11
−0.05
−0.04
−0.75
0.34
−0.21
0.17
−0.16
0.08
−0.57
−0.53
−0.27
−0.08
−0.06
Rank
bO
(1)
11.3
(2)
14288.4
(3)
−20.5
(4.5)
373.9
(4.5)
19.5
(6)
−41.9
(1)
44.8
(2)
−7.7
(3) −6801.0
(4)
−17.7
(5)
−205.4
(6)
−16.2
(1)
−1.0
(2)
−4.5
(3)
−383.3
(4)
−1.6
(5)
−23.4
(1) −32695.9
(2)
461.0
(3)
−23.4
(4)
68.6
(5)
−4.0
(6)
105.6
(1)
−143.6
(2)
−29.2
(3) −27215.3
(4)
−73.0
(5)
−926.5
Rank
(6)
(1)
(4)
(2)
(5)
(3)
(3)
(6)
(1)
(4)
(2)
(5)
(5)
(3)
(1)
(4)
(2)
(1)
(2)
(5)
(4)
(6)
(3)
(3)
(5)
(1)
(4)
(2)
For each investigated criteria, a high dependency on the pH-value is obvious. It is worth
noting that the entire possible range of pH-values for soils was investigated (3.0 ≤ pH ≤
7.5). The high pH dependency of DNDC was also identified by Stange (2001), who applied
a one-factor-at-a-time sensitivity analysis. As given by Eqs. 3.14 and 3.19, the pH-value
in SWAT-N controls both the ammonium-ammonia equilibrium, which determines the
consequent ammonia volatilisation, as well as the rate of nitrification. Higher pH-values
in the simulated soil lead to higher gaseous nitrogen emission, which is indicated by a b SD
of 0.66. In contrast, the nitrogen leaching is decreased (bSD of -0.52).
The parameter porous fraction from which anions are excluded (β) is positively related
to NL and negatively related to NE (Table 3.5). This can be explained since this parameter
51
controls the fraction of leachable nitrate in a soil layer.
3.6 Generalised likelihood uncertainty estimation (GLUE)
After the GSA, a generalised likelihood uncertainty estimation (GLUE, Beven & Binley,
1992) was conducted with the previously identified sensitive parameters. This was done
to further constrain the range of the identified parameters and to validate the new model
taking conditional uncertainty bounds into account. The principle idea of GLUE is that
several site specific models (i.e. model structure, parameters, input data, and observations)
produce acceptable representations of a systems output, which is sometimes referred to
as the phenomena of equifinality (Beven & Binley, 1992; Beven & Freer, 2001).
There are many similarities among GSA and GLUE such as the Monte-Carlo based
sampling strategy from predefined parameter distributions. In contrast to GSA, modelled output is compared with observations; in this study monthly nitrogen leaching. The
goodness-of-fit between the simulations of each site specific model (i.e. constant model
structure and input data; varying parameterisation) and observations are measured with
a likelihood function; and resulting models are a posteriori separated in non-acceptable
models with a likelihood of zero and acceptable models with a likelihood greater than
zero. Likelihood weighted conditional uncertainty intervals can than be computed for
each simulated time step. Hence, the GLUE methodology is a useful measure for model
uncertainty due to both parameter uncertainty and observation errors.
The sampling methodology for the GLUE analysis was the same as for the GSA, except
that 50,000 model runs were conducted and the value range for pH was set between 5.5
and 6.5, a typical pH-range for highly fertile loess soils in the Brandis area. The model
efficiency (NSE, Nash & Sutcliffe, 1970) was used as the likelihood function. A threshold
of N SE ≥ 0.35 was defined to select behavioural model runs. This relatively low value is
justified due to the model imperfections in predicting water percolation.
For the derivation of conditional uncertainty bounds, the likelihood measures were
rescaled in such a way, that the sum of likelihood yields one. Consequently, a probability
distribution of rescaled likelihood measures could be calculated. The behavioural model
runs were likelihood ranked in ascending order for each simulated time step, and the
distribution function of the predictions was calculated using the rescaled likelihood of the
corresponding model run. For details on the GLUE methodology, the reader is referred to
Beven & Freer (2001).
3.6.1 Results of the GLUE analysis
In Fig. 3.5 the parameter values providing behavioural model runs are presented. For a
complete definition of parameters the reader is referred to Table 3.6. Figures 3.5 B, E, and
F show that the parameters pH, EN O and ERB provide acceptable model runs over their
entire parameter range. Hence, the parameters EN O and ERB are relatively insensitive,
which is in agreement with the results of the previously employed GSA. The a-posteriori
parameter range of β and k1 (please note the logarithmic scale) are constrained (Fig. 3.5
A and D); only parameter sets including intermediate values of β and low values of k 1
52
3.7 Conclusion
lead to acceptable model runs.
It is worth to note that a visual interpretation of dotty plots according to a GLUE
analysis is as well a method to estimate the sensitivity of parameters (Beven & Freer,
2001), although it is subjective. However, the results of the GLUE analysis are consistent
with those obtained of the regression based GSA, except for pH, since a smaller range
was chosen for the GLUE analysis.
3.6.2 Model validation
The a-posteriori parameter sets were then used to calculate conditioned prediction intervals for each simulated time step. Figure 3.6 A shows the monthly observations of
N-leaching as well as the 95 % confidence level in accordance to the GLUE methodology.
SWAT-N simulated the seasonal cycle of nitrate leaching as well as the inter seasonal
variation accurately.
There are discrepancies between observed and predicted N-leaching at single months
such as in February 1987 and December 1991 (Fig. 3.6 A). The first one is caused by a
misprediction of water percolation (110 mm were observed but only 50 mm were simulated). Consequently, N-leaching was underpredicted by a factor of two. The deviation in
December 1991 can be attributed to an underestimation of simulated N-uptake by sugar
beet, which was the preceding crop type in winter 1991. Since there was more nitrate in
the simulated soil profile, the model overpredicted N-leaching.
SWAT-N underpredicts N-leaching at the beginning of the model period and over predicts it at the end of the period. This can be attributed to an imperfect representation
of atmospheric N-deposition rates. Observations at Brandis show a considerable decline
of N-deposition during the research period (Table 3.2), but the model does not include
deposition trends. A constant N concentration in rainfall of 7.9 mg N l −1 was calculated
as an input variable for the entire period, which led to an overestimation of N-leaching
during the 1980’s and an underestimation during the 1990’s.
It is worth to note that the implemented algorithms for nutrient cycling strongly enhanced the model performance (N SE of up to 0.4). Simulated average annual N L for
all behavioural model runs accounts for 42 kg N ha−1 a−1 , which is very close to the
observation of 47 kg N ha−1 a−1 . In contrast, the previous SWAT model failed to predict
NL at all (N SE =-0.22, see also Fig. 3.6 B). Simulated average annual N L with this
version accounted for 2 kg N ha−1 a−1 only, which is one order of magnitude lower than
the observation.
3.7 Conclusion
In this study we implemented algorithms from a processes oriented biogeochemical model
DNDC and a mechanistic denitrification function from CropSyst in SWAT. The new model
SWAT-N simulates decomposition, growth of nitrifying bacteria, nitrification, nitrificatory
as well as denitrificatory N emissions, N-uptake by plants and N transport due to water
fluxes.
53
0.39
0.6
0.8
1.0
5.6
5.8
6.0
β
pH
C
D
6.2
6.4
0.39
0.38
NSE
0.35
0.36
0.37
0.39
0.38
0.37
0.35
0.36
0.010
0.015
0.020
−4.5
−4.0
−3.5
−3.0
log10 k1
E
F
−2.5
−2.0
−1.5
0.39
0.38
0.36
0.35
0.35
0.36
0.37
NSE
0.38
0.39
0.40
kDN
0.40
0.005
0.37
NSE
0.38
0.36
0.35
0.4
0.40
0.2
0.40
0.0
NSE
0.37
NSE
0.38
0.37
0.35
0.36
NSE
0.39
0.40
B
0.40
A
0.16
0.18
0.20
ENO
0.22
0.24
0.55
0.60
0.65
0.70
0.75
0.80
ERB
Figure 3.5: Dotty plots according to the GLUE methodology applied to six sensitive parameters of SWAT-N focusing nitrogen leaching. The labels of the x-axis provide the
parameter range of the individual parameters. Sensitive parameters were derived
after a global sensitivity analysis. NSE means Nash & Sutcliffe efficiency; for the
definition of other parameters refer to Table 3.6. Please note the logarithmic scale
of k1 .
54
3.7 Conclusion
50
10
20
30
40
Observed
Upper bound
Lower bound
0
A
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
50
10
20
30
40
Observed
Simulated
0
B
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
Time (month)
Figure 3.6: A: Results of observed monthly N-leaching (arithmetic mean of three replicates)
at lysimeter station Brandis, Germany, (Haferkorn, 2000) and conditional 95 %
uncertainty interval according to the GLUE methodology simulated with SWAT-N.
B: Results of observed monthly N-leaching and best simulation run of the previous
SWAT version.
55
It can be concluded from the regression based GSA, that the rate coefficient for decomposition (k1 ), pH-value, and the porous fraction from which anions are excluded are the
most important parameters controlling N turnover and N-leaching.
The implemented algorithms improved the models capability to predict mineral Nleaching, since model efficiencies of up to 0.4 could be achieved. However, considerable
discrepancies between simulated N-leaching and observed N-leaching exists for individual
years. This can be referred to both inaccuracies of the implemented models for crop
growth and N-uptake. Furthermore, the restriction of constant N concentration in rainfall
limits the applicability of SWAT-N for long term investigations, when a clear trend of
atmospheric N-deposition is visible.
In this study, the new conceptualisation of SWAT-N was tested against lysimeter data,
because the focus was on the evaluation of those processes that take place on the land
phase. Clearly, the new model SWAT-N needs to be additionally tested on the catchment
scale, prior an application for the assessment of landuse on water quality on that scale.
This will be done in a future work.
This study has been supported by the Deutsche Forschungsgemeinschaft within the scope
of the Sonderforschungsbereich 299 (SFB299). We are indebted to Dr. Ulrike Haferkorn
and Dr. Uwe Franko (both UFZ-Leipzig, Germany) for the provision of lysimeter data
from the research station Brandis.
56
Table 3.6: Parameters and their range for the employed sensitivity analysis.
Parameter Lower limit Upper limit Definition
k1
2.5 × 10−5
2.5 × 10−3 Rate of decomposition
of very labile organic matter
kNI
1.25 × 10+1 5.0 × 10+1 Rate of nitrification
kDN
5.0 × 10−3
2.0 × 10−2 Rate of denitrification
−1
kRB
1.65 × 10
6.6 × 10−1 Rate of decomposition
of fast decomposable microbes
−2
−2
khrb
2.0 × 10
8.0 × 10
Rate of decomposition
of slow decomposable microbes
−1
−1
ENO
1.6 × 10
2.4 × 10
Efficiency coefficient
for resistant humads decomposition
ERB
5.36 × 10−1 8.04 × 10−1 Efficiency of microbes
EAC
1.6 × 10−1
2.4 × 10−1 Efficiency coefficient
for microbes decomposition
−2
−2
DNH3
2.0 × 10
3.0 × 10
Ammonia diffusion coefficient
−3
−3
DRF
1.08 × 10
1.62 × 10
Depth related parameter
β
0.0 × 100
1.0 × 100
Porous fraction
from which anions are excluded
0
0
pH
3.0 × 10
7.5 × 10
pH-value
57
Table 3.7: Values of parameters which remained fixed for the GSA and GLUE analysis as well
Symbol
k N2 O
kNO
CLMX
CN1
CN2
CN3
CNB
CNH
TMX
TOPT
TA
TMXNI
TOPNI
TANI
γ
MCRIT
58
as their sources.
Definition
Coefficient for N2 O
production
Coefficient for NO
production
Maximum clay content
for clay adsorption
C/N ratio of very fast decomposable
organic litter (OL)
C/N ratio of fast decomposable OL
C/N ratio of slow decomposable OL
Microbial C/N ratio
C/N ratio of humads
Maximum temperature
for decomposition
Optimum temperature
for decomposition
Scale parameter
for decomposition
Maximum temperature
for nitrification
Optimum temperature
for nitrification
Scale parameter
for nitrification
Scale parameter
for decomposition
Half optimum soil
moisture for decomposition
Value
6 × 104 d−1 WFPS−1
Source
Zhang et al. (2002)
2.4 × 10−3 d−1 WFPS−1
Zhang et al. (2002)
63 %
5.0
DNDC source code
50.0
200.0
10.0
10.0
58.00
34.15
Stange (2001)
45.00
Own estimation
33.33
1.90
DNDC source code
DNDC source code
DNDC source code
DNDC source code
Stange (2001)
Stange (2001)
5.39
Li et al. (1992)
Own estimation
Own estimation
0.15 WFPS
Stange (2001)
0.425 WFPS
Stange (2001)
4 Pohlert et al., 2006, Hydrol. Earth
Syst. Sci. Discuss.,3, 2813–2851.
Der folgende Text erschien im Jahre 2006 in der Zeitschrift Hydrology and Earth
System Sciences Discussion. Das Copywright liegt bei den Verfassern des Textes. Eine
überarbeitete Versions dieses Textes unter Berücksichtigung der Kommentare des Open
Review Prozesses wurde für den Band Hydrology and Earth System Sciences eingereicht.
T. Pohlert, Breuer, L., Huisman, J. A., Frede, H.-G., 2006. Assessing the
model performance of an integrated hydrological and biogeochemical model
for discharge and nitrate load predictions. Hydrol. Earth Syst. Sci. Discuss.,
3, 2813–2851.
Abstract. In this study, we evaluate the performance of the SWAT-N model, a
modified version of the widely used SWAT version, for discharge and nitrate predictions at the mesoscale Dill catchment for a 5-year period. The underlying question
is, whether the model efficiency is sufficient for scenario analysis of land-use changes
on both water quantity and quality. The Shuffled Complex Evolution (SCE-UA) algorithm is used to calibrate the model for daily discharge at the catchments outlet.
Model performance is assessed with a split-sampling as well as a proxy-basin test
using recorded hydrographs of four additional gauges located within the catchment.
The efficiency regarding nitrate load simulation is assessed without further calibration on a daily, log-daily, weekly, and monthly basis as compared to observations
derived from an intensive sampling campaign conducted at the catchments outlet.
A new approach is employed to test the spatial consistency of the model, where
simulated longitudinal profiles of nitrate concentrations were compared with observed longitudinal profiles. It is concluded that the model efficiency of SWAT-N is
sufficient for the assessment of scenarios for daily discharge predictions. SWAT-N
can be employed without further calibration for nitrate load simulations on both
a weekly and monthly basis with an acceptable degree of accuracy. However, the
model efficiency for daily nitrate load is insufficient, which can be attributed to
both data uncertainty (i.e. point-source effluents and actual farming practise) as
well as structural errors. The simulated longitudinal profiles meet the observations
reasonably well, which suggests that the model is spatially consistent.
Keywords. SWAT-N; calibration; validation; nitrogen cycle; catchment model
4.1 Introduction
In the last decades the Dill catchment, which is located in Mid-Hesse, Germany, faces
a decline in farming, which has led to an increase of fallow land. Currently, fallow land
59
contributes to about 9% of the area. This change in land-use has severe consequences for
various landscape functions, which led to the foundation of the Collaborative Research
Centre (Sonderforschungsbereich 299, SFB 299) at the Justus-Liebig-University Gießen.
The scope of the SFB 299 is to develop and evaluate both economically and environmentally sustainable land-use options for peripheral regions, such as the Dill catchment.
Within the framework of the SFB 299, various models coming from the fields of economy,
hydrology, soil science, and biology are used to assess the impact of potential land-uses on
various landscape functions, such as food production, social welfare, habitat for plant and
animal species, tourism and leisure, flood protection and water harvest (Frede & Bach,
1999). As members of the SFB 299, our aim is to evaluate potential land-use options in
regard to water balance and alteration of river water quality on the landscape-scale, which
can be achieved by using eco-hydrologic models.
During the last 20 years, various nitrogen turn-over and transport models were developed for the plot and field scale (i.e. no spatial variability of site parameters) to simulate
processes within agro-ecosystems. Historically, these models were developed by both crop
scientists and soil scientists. Although both groups work on the same research object
(i.e. agro-ecosystem), the focus (e.g. crop productivity, groundwater contamination) is
different. Consequently, processes that occur within agro-ecosystems (e.g. crop-growth,
soil water movement, nutrient cycle and transport) are modelled with a varying degree of
detail depending on the focus.
Algorithms to simulate nitrogen cycle and transport are incorporated in crop-growth
models that were developed by crop scientists, such as the WOFOST model (Supit et al.,
1994), the CropSyst model (Stöckle et al., 2003), and models of the CERES-family (Jones
& Kiniry, 1986). Agro-ecosystem models developed by soil scientists focus on environmental risks due to farming, such as nitrate leaching into groundwater (e.g. HERMES,
Kersebaum, 1995) and gaseous nitrous emissions into the atmosphere (Li et al., 1992).
A few agro-ecosystem models were build to investigate both crop productivity and environmental pollution due to fertilisation on the plot-scale. Two examples are the EPIC
model (Williams et al., 1984) and Crop-DNDC (Zhang et al., 2002), which both have
incorporated concepts of the aforementioned models.
Although the modelling of water and nutrient fluxes at the plot scale is well-represented,
up-scaling of the concepts underpinning these models to provide predictions at the
landscape-scale has been identified as an area of needed research (Sivapalan et al., 2003).
The concepts within eco-hydrologic models that simulate both water and nitrogen cycle
on the catchment scale range from simple empirically based models to complex processoriented models. Simple empirically based methods were developed for large-area applications such as source-apportionment methods (e.g. Grizetti et al., 2005), data-based
nutrient retention models for various aquatic systems (EUROHARP-NUTRET, Kronvang
et al., 2004), and GIS-aided static approaches, which include population equivalents for
the estimation of nitrogen input in river systems from point-sources and diffuse emissions based on nitrogen surplus in soils, such as MONERIS (Behrendt et al., 2000) or the
PolFlow model(de Wit, 2001). The main advantages of the aforementioned methods are
their ease of use and their low computational demand. However, their use for scenarioanalysis is questionable due to several incorporated simplifications and assumptions.
The process-oriented models for the application on large scale catchments have in-
60
4.1 Introduction
corporated concepts of crop-growth and nutrient cycle models. Examples are the semidistributed HBV-N model (Bergström, 1995; Arheimer & Brandt, 1998) and LASCAM
model (Sivapalan et al., 2002; Viney et al., 2000), which both simulate water and nitrate
dynamics on the land-phase within subbasins. A strategy to reflect the spatial variability
of soil, land-use, and farming practise within subbasins are Hydrotopes or Hydrologic Response Units (HRU). Examples for models with this concept are the SWAT model (Arnold
et al., 1998), and its spin-off – the SWIM model (Krysanova & Haberlandt, 2002).
Fully distributed models were developed in the hope that an improved representation
of spatial variability of land-use and soil characteristics will lead to an improved model
performance regarding to discharge and river pollution prediction. Examples for this type
of models are the INCA model (Whitehead et al., 1998a; Wade et al., 2002), which requires hydrologic effective rainfall calculated with an external water balance model as
a driving input variable, the TNT model (Beaujouan et al., 2001) that is build on the
TOPMODEL approach (Beven et al., 1995), and MIKE SHE (Refsgaard & Storm, 1995).
Although fully distributed models are believed to be the most ’realistic’ models to represent the hydrologic system, inevitable constraints are their demands for both input
data and computational power. Hence, these models can often only be employed for the
simulation of small watersheds.
In this study we use the semi-distributed eco-hydrologic model SWAT-N (Pohlert et al.,
2006a) to assess the impact of potential land use on both the hydrologic cycle as well as
the nitrogen cycle, respectively. The hydrologic part of the SWAT-N model is based on
the Soil and Water Assessment Tool (SWAT, Arnold et al., 1998). The modifications
regarding runoff generation in mountainous regions done by Eckhardt et al. (2002) are
incorporated into SWAT-N. The conceptualisation of the nitrogen cycle in SWAT-N is
based on the Denitrification-Decomposition model (DNDC, Li et al., 1992, 2000) as well
as on the CropSyst model (Stöckle et al., 2003).
We employed an extended approach for model testing. As Kuczera & Franks (2002)
critised, many simulation studies compare simulated with observed discharge at the final
outlet of a catchment only, and conclude that the employed model is “validated”, although
other catchment responses remained unconsidered in the model testing. In this study we
employed conventional split-sampling tests as well as proxy-basins tests (Klemeš, 1986) to
measure the accuracy and spatial consistency of predicted discharge (five gauges) as well as
nitrate load (two gauges). Furthermore, we compare observed and simulated longitudinal
nitrate concentration profiles (Grayson et al., 1997) taken under various flow conditions.
This new approach allows the identification of the correct representation of spatially
distributed sources of nitrate as well as the routing of nitrate along the main channel.
The specific research questions that we would like to address in this contribution are:
1. Is the SWAT-N model able to predict daily river discharge at multiple gauges within
the Dill catchment at the same time?
2. Does the SWAT-N model accurately predict daily, weekly, and monthly nitrate loads
at the catchments outlet without further calibration of parameters that govern the
nitrogen cycle?
3. Does SWAT-N accurately predict the longitudinal profile of nitrate concentrations
61
along the main channel and nitrate concentrations of tributaries at stable flow conditions?
62
4.2 Materials and methods
4.2.1 The Dill catchment
The Dill catchment (692 km2 ) is located in the mountainous area of Hesse, Germany,
with elevation ranging between 155 to 675 m.a.s.l. and a mean slope of 8 according to
the digital elevation model. Approximately 55% of the catchment is forested and both
arable and pasture land contributes to about 28% of the basin based on remotely sensed
data (LANDSAT TM5) taken in 1994/95 (Nöhles, 2000). The land-use distribution within
the basin is shown in Fig. 4.1.
AGRL
PAST
CVEG
FRSD
FRSE
URBN
WATR
0
Units (m) 10000
Figure 4.1: Map of actual land-use in the Dill catchment based on LANDSAT TM5 data
(Nöhles, 2000). AGRL means arable land, PAST is pasture, CVEG is changing
vegetation, FRSD is deciduous forest, FRSE is evergreen forest, URBN means urban
area, and WATR denotes surface water.
The geology in the south of the catchment comprises of Carboniferous clay schist and
graywacke, the centre of the basins is dominated by both Devonian volcanic rocks and
graywacke, and in the northern part Devonian quartzite sandstones can be found. The
rocks are widely covered by periglacial deposits, which resulted from solifluction during
the Pleistocene. This process lead to a considerable anisotropy of vertical and horizontal
saturated hydraulic conductivity due to coarse fragments, which are aligned with their
longitudinal axis in parallel to hill slopes (Eckhardt et al., 2002; Fröhlich et al., 2005). In
river valleys, colluvium with larger depths developed during the Holocene. The dominant
soil type throughout the catchment is a shallow cambisol with underlying rocks with a
low permeability (Sauer, 2002).
63
Annual average air temperature is 9.5 C and annual average areal precipitation is 856
mm for the period under investigation (1.7.2000 to 31.12.2004). Discharge at the river
outlet Aßlar accounts for 412 mm.
4.2.2 Spatial input data
For this study, a 25 × 25 m2 digital elevation model (DEM) and a vectorised soil type
map (1:50,000) including soil-profile-data were available (BFD50, Hessisches Landesamt
für Umwelt und Geologie, 2000). Since pH-values are not included in the BFD50 data
set hitherto, they were taken from Sauer (2002) with 5.9, 4.1, 5.1, and 4.9 for arable
land, forests, pasture land, and fallow land in the Dill catchment, respectively. A landuse map was compiled from multi-temporal LANDSAT TM5 images taken in 1994/95
(Nöhles, 2000). Further processing of the maps, such as the delineation of watersheds,
the HRU distribution, and the interpolation of ’point’ rainfall at precipitation stations
to each subbasion (nearest-neighbour-approach) was accomplished within the AVSWAT
preprocessing-tool (Di Luzio et al., 2002).
4.2.3 Nitrogen input data
Data on crop rotation were taken from Lenhart (2003), where a crop rotation of winter
rape - winter barley - oat was assumed to be typical for arable land in the Dill catchment.
Fertilisation rates for each crop are 145 kg N ha−1 , 50 kg N ha−1 , and 50 kg N ha−1 ,
respectively. A fertilisation rate of 55 kg N ha−1 was specified for pasture land.
Data of average monthly point source effluents were available from three municipal
sewage treatment plants as well as from one steel mill, which uses nitric acid to harden
the steel. The nitrate rich wastewater of the steel mill is released into the Dietzhölze (see
Fig. 4.2 a). The effluents of these four point sources comprise about 90 % of total nitrate
released by point sources in the entire catchment (Lenhart, 2003). The average monthly
data of point source effluents were assumed to be representative for the entire simulation
period.
A constant nitrate concentration in precipitation of 1.5 mg N l −1 was specified for the
model, which was calculated from annual average precipitation and reported atmospheric
nitrogen deposition (Gauger et al., 2001).
4.2.4 Hydrologic and meteorological data
The Dill catchment is monitored by five river gauges, which are operated by the Hessisches
Landesamt für Umwelt und Geologie (HLUG). The location of the river gauges as well as
their drained subbasins are given by Fig. 4.2 a. For this study, daily discharge data for
the period 1.7.2000 - 31.12.2004 were used for the assessment of the model.
Furthermore, records of daily precipitation of 12 stations from the precipitation network
of the German Weather Service were used. The meteorological data set was completed
with records of daily maximum and minimum temperature, wind speed, air humidity, and
sun shine duration of two climatic stations.
64
a)
b)
Dietzhölze
Obere Dill
7
1
3
12
2
18
Aar
4
Dill
27
28
33
34
36
35
38
39
43
0
Units (m) 10000
River gauges
5
1 Dietzhölze 3 Obere Dill 5 Aßlar (Outlet)
2 Dillenburg
0
Units (m) 10000
50
51 53 56
57
Sampling sites
4 Aar
Figure 4.2: Map of the Dill catchment and its gauged subbasins (left). Locations for snapshotsampling along the river Dill and its tributaries(right).
Three of the 12 precipitation stations were abandoned during the research period. The
series of these precipitation stations were extended with estimated daily precipitation
based on multiple linear regression models using the records of the surrounding stations as
predictor variables. Cross-validation provided root-mean-squared-errors (RMSE) between
1.63 to 2.20 mm.
The SWAT-N model will require daily global radiation records if the Penman-Monteith
method is used for the calculation of evapotranspiration. Therefore sun shine duration
was converted to daily global radiation. Hereby, the relation as proposed by Angstrøm
(1924) was used, where the ratio of global radiation and extra-terrestrial radiation equals
the proportion of hours of bright sunshine for a given location and day of the year.
4.2.5 Hydrochemical Data
From April 2000 until December 2005, an automated sampling station (ISCO 3700, ISCO,
Lincoln, NE, USA) at the gauge Aßlar (see Fig. 4.2 a) took hourly time-proportional
mixed samples. These samples were collected twice a week bulked to yield a composite
daily sample, filtered through 0.45 µm polypropylene-membrane filters (Whatman puredisc, Whatman Inc., Clifton, NJ, USA) and stored at -20 C until the water samples
were chemically analysed. For the period of April 2000 to April 2002, an automated
photometric method (Technicon Autoanalyzer N, Technicon Industrial Systems, Tarrytown, NY, USA) was used to measure nitrate. Nitrate is hereby reduced to nitrite by
Cu-II-sulphate, colourised to yield a diazo dye, and photomotrically measured at 520 nm.
From May 2002 onwards, the samples were analysed with an ion chromatograph (Dionex
DX-120, Dionex Corp., Sunnyvale, CA, USA) in accordance to DIN EN ISO 10304-1
(1995). To account for the systematic difference between the two chemical analysis meth-
65
ods, a simple linear regression model was employed to homogenise the measured nitrate
concentration series. The homogenised nitrate series refer to the photometric method
(y = 0.91 + 0.83x, R2 = 0.72).
Highest nitrate concentrations at Aßlar are measured in summer due to a low capacity
for the dilution of point source effluents. The 10% percentile, median and 90% percentile
for nitrate concentrations at gauge Aßlar are Q10 = 1.8 mg N l−1 , Q50 = 3.2 mg N l−1 ,
and Q90 = 5.9 mg N l−1 , respectively.
Furthermore, values of instantaneous nitrate samples, which were usually collected on
a monthly basis at gauge Obere Dill (2000 - 2003, n = 37), were taken from the on-line
publication of the Hessische Gewässergütebericht (www.hlug.de). In contrast to gauge
Aßlar, the highest concentrations at this up-stream gauge are present during wintertime,
which indicates the absence of major point sources. The nitrate concentrations are lower
as compared to Aßlar with Q10 = 0.4 mg N l−1 , Q50 = 0.6 mg N l−1 , and Q90 = 0.9 mg
N l−1 .
Additionally, four longitudinal profiles were sampled during two low flow conditions,
one high flow and one intermediate flow condition to check the internal consistency of the
model with regard to the geographical sources of the nitrate emissions. The samples were
also analysed for nitrate with the ion chromatograph. The sampling sites at the main
channel and at the tributaries are depicted in Fig. 4.2 b.
4.2.6 The SWAT-N model
The SWAT-N model (N for Nitrogen) is a modified version of the SWAT (Arnold et al.,
1998) and SWAT-G model (Eckhardt et al., 2002) in which the conceptualisation of the
nitrogen cycle and transport has been changed. It is a conceptually based, though process
oriented semi-distributed eco-hydrologic model, which operates on a daily time step.
The smallest spatial modelling unit in SWAT-N is a hydrological response unit (HRU),
which is a unique combination of land-use type and soil type derived through the overlay
of a soil and a land-use map. Processes such as surface runoff estimated with the SCS curve
number equation, percolation with a layered storage routing technique, lateral subsurface
flow, potential evapotranspiration by the Penman-Monteith method, snow melt, crop
growth, soil erosion, nitrogen and phosphorous cycle are simulated for each HRU (Arnold
& Fohrer, 2005).
Based on a DEM, the catchment is delineated into several subbasins and a topological
stream network is derived. The pre-processing of GIS data was done with the AVSWAT
tool (Di Luzio et al., 2002). The resulting water, sediment, and nutrients fluxes are seperately summarised for each HRU within a subbasin, and allocated to the main channel
of the particular subbasin. Routing of water, sediment, and solutes within the stream
network is calculated with the variable storage routing method (Arnold et al., 1995).
The algorithms for in-stream water quality processes such as nitrate uptake by algae are
incorporated from the QUAL2E model (Brown & Barnwell, 1987).
Eckhardt et al. (2002) modified the SWAT99.2 version to yield high proportions of
interflow as compared to other flow components. Interflow is the dominant flow component in mountainous regions with shallow soils, developed from periglacial layers above
rocks with low permeability. The SWAT-G model (Eckhardt et al., 2002) considers the
66
anisotropy of vertical and lateral hydraulic conductivity as well as the low permeability
of the underlying bedrock, which is present in vast areas of the Dill catchment.
The original conceptualisation of the nitrogen cycle in both SWAT and SWAT-G is
based on the EPIC-model (Williams et al., 1984). However, the EPIC based SWAT model
failed to predict N-cycle reasonably, since high denitrification losses of up to 135 kg N ha −1
yr−1 were simulated for single HRU’s within the Dill catchment. This can be explained
by the conceptualisation of denitrification in SWAT (Neitsch et al., 2002). Denitrification
occurs, whenever 95% of field capacity is exceeded. Since water will only percolate in the
model if soil moisture exceeds field capacity, denitrification and nitrogen leaching are two
heavily competing processes in the EPIC-based SWAT versions. Under humid climatic
conditions, where soils are moist for extended periods of the year, the EPIC approach
leads to a rapid and complete depletion of the simulated nitrate pools in each layer due
to denitrification (Pohlert et al., 2005c).
To overcome the aforementioned limitations, algorithms from both the DNDC (Li et al.,
1992, 2000) and the CropSyst (Stöckle et al., 2003) model were used to replace the EPIC
approach for nitrogen simulation in SWAT. The new version, which includes the modifications made in SWAT-G and the algorithms for nitrogen cycling, is further denoted
as SWAT-N. It simulates decomposition of organic matter from three different organic
pools with first-order-kinetics, microbial immobilisation of decomposed organic nitrogen
as well as mineral nitrogen, clay adsorption of ammonium, ammonium – ammonia equilibrium, ammonia volatilisation, nitrification including nitrificatory nitrogen emissions to
the atmosphere, and denitrification. A detailed description of the conceptualisation of the
nitrogen cycle within SWAT-N as well as the sensitivity of model parameters are given
by Pohlert et al. (2006a).
4.2.7 Calibration
With the term ’model calibration’ we mean the adjustment of model parameters within
physically reasonable spans to yield a better match between an observed and simulated
variable for a specific time period. Only the most sensitive parameters with regard to the
simulated hydrologic cycle were calibrated in this study. The parameters, which control
N-cycle, were manually set after an extensive sensitivity analysis (Pohlert et al., 2006a).
In this study, a split-sampling-test (Klemeš, 1986) was conducted using the recorded
hydrograph at gauge Aßlar for the period from 1.7.2000 - 31.1.2003 and the period 1.2.2003
- 31.12.2004 for calibration and model assessment, respectively. Additionally, a proxybasin test (Klemeš, 1986) was employed, where simulated hydrographs were compared
with observations of four interior gauges (Fig. 4.2 a) using the same parametrisation as
for the gauge Aßlar. A two years warm-up period was used to initialise the pools of the
model.
A single criteria calibration was conducted with the Shuffled Complex Evolution Algorithm developed at the University of Arizona (SCE-UA, Duan et al., 1992) for minimising
the sum-of-squared-residuals between the simulated hydrograph and observations. The
SCE-UA method is an optimisation algorithm, which searches for the ’global optimum’
within a given parameter space based on a random search strategy combined with a downhill simplex method. SCE-UA has been successfully applied for hydrologic calibration (e.g.
67
Hogue et al., 2000; Duan et al., 1992; Eckhardt & Arnold, 2001). Table 4.1 provides the
parameters and their upper and lower bounds, which were selected for calibration.
Only six parameters were selected for calibration in this study, whereby four of these
parameters vary in space (i.e. refer to HRU’s). Since the number of HRU’s in the Dill
model is very high (795 HRU’s), an independent calibration of these parameters is neither
meaningful nor feasible. Therefore, the calibration methodology as proposed by Grayson
& Blöschl (2000) and Eckhardt & Arnold (2001) was applied. In this methodology a
pattern of a spatially distributed parameter is calibrated by varying a single parameter
that sets the values for the whole pattern according to predefined ratios.
4.2.8 Assessment of model performance
Various goodness-of-fit measures were calculated, and both graphs and thematic maps
were visually analysed for the assessment of model efficiency to predict both discharge
and nitrate load as recommended by Legates & McCabe (1999).
The model efficiency (E) after Nash & Sutcliffe (1970) was calculated for daily observed
and simulated discharge, where
E =1−
n
X
i=1
2
(Oi − Pi ) /
n
X
i=1
Oi − Ō
2
(4.1)
Oi and Pi are observed and predicted values, respectively, and Ō is the observed average.
E ranges between 1 and −∞, where E = 1 denotes a ’perfect’ model fit and E = 0 means
that the value for average observed discharge is as good a predictor as the hydrologic
model.
A disadvantage of E is the fact that due to the squaring of residuals, E is highly sensitive
to mis-predicted flood peaks (Legates & McCabe, 1999), and is almost insensitive to mismatches during low flows. Since low flows are of great importance for the capacity of the
river to dilute point source effluents, where relatively small residuals of predicted flows can
result in large discrepancies between simulated and observed nitrate concentrations, a logtransformation of both observations and simulations was applied. The model efficiency
of the log-transformed data (Elog ) was calculated using the above given equation. The
log-transformation emphasises the differences between observed and simulated low flows
as well as recession curves, whereas the impact of flood peaks is less emphasised since
they are flattened . E was additionally calculated for both weekly and monthly averaged
discharge.
To assess the model performance in regard to nitrate predictions at the gauge Aßlar and
Obere Dill, the index of agreement (D) according to Wilmott et al. (1985) was additionally
calculated, whith
D =1−
n
X
i=1
| P i − O i |2 /
n
X
i=1
| Pi − Ō | + | Oi − Ō |
2
(4.2)
The index of agreement ranges between 0 and 1. It should be noted that the values of
the measures E and D are not directly comparable, though both measures are correlated
due to the ’squared residuals’ term, which is apparent in both numerators of the equations.
68
4.3 Results and Discussions
Table 4.1: Calibrated parameters for the Dill catchment (1.7.2000 - 31.1.2003). The soil-unit
2458 refers to a shallow soil that is present on both upper slopes and the top of the
hills. Soil-unit 202 is typical for lower slope positions. Spatially distributed parameters were calibrated as proposed by Eckhardt & Arnold (2001).
Parameter
Baseflow recession constant (d−1 )
Groundwater delay time
for aquifer reacharge (d)
Bulk density moist,
soil-unit 2458, layer 4 (g cm−3 )
Anisotropy, soil-unit 2458, layer 3
Lower limit
0.03
1
Upper limit
0.06
20
Calibrated value
0.034
19.8
2.51
2.64
2.64
10
85
83.3
1
45
44.9
2
8
4.65
For visual inspection, comparative plots of measured and simulated nitrate concentrations for 11 sites along the main channel and six tributaries sampled at four different dates
were drawn. The post-processing tool hru2map (Pohlert, 2005) was used to re-assemble
model output for HRU’s to a grid map, by overlaying the subbasin map, land-use map as
well as the soil map as proposed by Haverkamp et al. (2005). The site-specific visualisation
of hydrologic variables aids for error identification in the model (Haverkamp et al., 2005).
4.3.1 Discharge
Table 4.1 gives the upper and lower bounds as well as the final calibrated values for the
parameters considered for the hydrologic calibration. The calibrated values are similar
and consistent as compared to those reported by previous studies conducted in the Dill
catchment (e.g. Huisman et al., 2004; Eckhardt et al., 2005), although different calibration
periods were used in these studies. This confirms both the robustness of the SWAT code
as well as the SCE-UA calibration algorithm.
Table 4.2 provides the goodness-of-fit measures for simulated and observed discharge for
each gauge within the Dill catchment. As explained in section 4.2.7, only the hydrograph at
gauge Aßlar was considered for the single-criteria calibration. For the calibration period
the efficiency measure E for daily data is lower at the gauges of the head catchments
(Diethölze, Obere Dill, Aar) as for those, which drain larger proportions of the catchment
(Aßlar, Dillenburg). The low flows at Aßlar and Dillenburg are well simulated, which is
indicated by E = 0.85 and E = 0.84 for log-transformed values, respectively. In contrast,
the predictions for low flows at the gauges Aar and Obere Dill are less accurate, as
indicated by E = 0.14 for both gauges.
The poor model performance at the head-catchments for low flows can be explained
69
Table 4.2: Goodness-of-fit measures for simulated and observed discharge at various gauges
for the calibration period (1.7.2000 - 31.1.2003) and assessment period (1.2.2003 31.12.2004). E denotes model efficiency after Nash & Sutcliffe (1970), the indices d,
log, w, and m denote daily data, log-transformed data, weekly averages, and monthly
averages, respectively.
Calibration
Assessment
Ed
Elog
Ew
Em
Ed
Elog
Ew
Em
Aßlar
0.72
0.85
0.93
0.96
0.63
0.80
0.86
0.87
Dillenburg
0.76
0.84
0.92
0.97
0.74
0.74
0.89
0.92
Dietzhölze
0.66
0.71
0.89
0.91
0.56
0.60
0.81
0.84
Aar
0.66
0.14
0.79
0.80
0.68
-0.43
0.88
0.90
Obere Dill
0.51
0.14
0.67
0.74
0.66
0.36
0.87
0.94
by the fact that the geology in these areas (Devonian quartzite sandstones) differ as
compared to the rocks of the remaining catchment areas. A single-criteria calibration
focusing on discharge at the catchments outlet, as it has been done in this study, leads to
an ’optimised’ integral parameterisation of groundwater related parameters (i.e. ’baseflow
recession constant’, and ’groundwater delay time for aquifer recharge’) for the entire
catchment, which are not representative for the head-catchments.
As expected, the goodness-of-fit increases for both weekly and monthly averaged discharge data, which can be explained by the effect of smoothing. The model efficiencies
slightly decline for the assessment period for all gauges except for the Obere Dill gauge.
It is noteworthy that Huisman et al. (2004) found model efficiencies of E = 0.85 and
E = 0.8 for the calibration (1986 - 1988) and validation period (1989 - 1991) for the
gauge Aßlar, respectively, which suggests a ’better’ model performance as compared to
this study. As mentioned previously, the ’optimal’ parameterisation in their work was very
similar to the parameterisation found in the current study. This illustrates the importance
of the selected simulation period with regard to model performance, since weather patterns
as well as the quality of weather data (three rain gauges went out of service in our study
period), have a strong impact on the accuracy of simulated hydrologic processes.
Scatterplots of simulated and observed discharge for the entire simulation period at the
gauges Obere Dill and Dillenburg are depicted in Fig. 4.3. The larger scatter for daily
data in the plot for gauge Obere Dill (Fig. 4.3 a) as compared to the gauge Dillenburg
(Fig. 4.3 c) agrees with the model efficiencies for daily data presented in Table 4.2. It
is obvious that SWAT-N underestimates low flows at the gauge Obere Dill (Fig. 4.3 b),
whereas low flows at the gauge Dillenburg (Fig. 4.3 d) are more accurately predicted.
The average groundwater contribution (2000-2004) to the reach for each delineated
subbasin is given in Fig. 4.4. The simulated average annual groundwater contribution to
the reach for the catchments Dietzhölze (73 mm), Aar (33 mm), Dillenburg (88 mm),
Haiger (76 mm) and Aßlar (76 mm) fall within the range of reported annual groundwater contributions based on various estimation methods after Kaviany (1978). The high
70
0
5
10
15
20
3
1.5
0
−1.5
−3
Observed discharge (m3 s−1)
50
Observed discharge (m3 s−1)
0
1.5
3
75
0
1
2
3
Dillenburg
−1
50
25
Simulated discharge (log10(m3 s−1))
d)
75
Dillenburg
25
0
−1.5
Observed discharge (log10(m3 s−1))
0
Simulated discharge (m3 s−1)
c)
Upper Dill
−3
Simulated discharge (log10(m3 s−1))
5
10
15
20
b)
0
Simulated discharge (m3 s−1)
a) Upper Dill
−1
0
1
2
3
Observed discharge (log10(m3 s−1))
Figure 4.3: Scatterplots of simulated and observed discharge at gauges Obere Dill and Dillenburg. Daily discharge (left) and log-transformed daily discharge (right) are depicted
for the period 1.7.2000 - 31.12.2004. The dashed line indicates the one-to-one line.
71
250
200
Dietzhölze
150
100
Obere Dill
50
Aar
Dill
0
Units (m) 10000
Figure 4.4: Map of simulated annual average groundwater contribution to the reach of each
subbasin (2000-2004).
groundwater contribution in the south-west of the Dill catchment can be explained by the
fact that this area receives more precipitation (station Driedorf 1230 mm) as compared
to the aerial average precipitation of the catchment with 856 mm, and the hillslope in the
south-east is lesser than throughout the catchment.
4.3.2 Nitrate loads
The performance of SWAT-N with regard to nitrate load predictions (1.7.2000-31.12.2004)
at the gauges Aßlar and Obere Dill is summarised in Table 4.3. It it worth noting that
the parameters, which control N-cycling within SWAT-N, were not calibrated. Hence, the
presented results can be directly interpreted for model assessment.
The SWAT-N model overestimates average daily nitrate load (N̄o ) at gauge Aßlar and
Obere Dill by ca. +13% and +77%, respectively. However, it should be noted that only 37
instantaneous nitrate samples were available for the gauge Obere Dill to calculate average
daily nitrate load.
The simulated and observed nitrate load for gauge Aßlar is shown in Fig. 4.5. SWATN simulates the seasonal cycle of nitrate load accurately, which can be taken from the
graphs of monthly (Fig. 4.5 c) and weekly (Fig. 4.5 b) loads, respectively. At week 90
and 140 SWAT-N under- and overestimates the average weekly nitrate loads, respectively
(Fig. 4.5 b). The daily variability of nitrate loads are accurately simulated (Fig. 4.5 a,
and Table 4.3), though the timing as well as the absolute amount of nitrate load peaks
are mis-predicted.
72
2 × 104
1 × 104
0
NO3 load (kg N d−1)
3 × 104
a) Daily
1.7.00
1.7.01
1.7.02
1.7.03
1.7.04
Date
5 × 103
1.5 × 104
Observed
Simulated
0
2.5 × 104
b) Weekly
0
50
100
150
200
Weeks after 1.7.2000
2.5 × 103 5 × 103 7.5 × 103 1 × 104
0
c) Monthly
Jul 00
Jul 01
Jul 02
Jul 03
Jul 04
Month
Figure 4.5: Simulated and observed nitrate load at gauge Aßlar (1.7.2000 - 31.12.2004).
73
Table 4.3: Goodness-of-fit measures for nitrate load predictions (1.7.2000-31.12.2004) at gauges
Aßlar and Obere Dill. E means model efficiency (Nash & Sutcliffe, 1970), D is the
index of agreement (Wilmott et al., 1985), N̄ is average nitrate load (kg N d−1 ), sd is
standard deviation (kg N d−1 ), n is the number of data pairs, and the indices o and
p mean observed and simulated, respectively. Only monthly instantaneous samples
were available for the Obere Dill.
Gauge
Aßlar
Obere Dill
Values
daily
log daily
weekly
monthly
daily
log daily
E
-0.29
0.25
0.22
0.65
-3.44
0.07
D
0.62
0.73
0.81
0.89
0.64
0.87
N̄o
2247
7.4
2231
2397
48
1.4
N̄p
2527
7.6
2523
2502
85
1.2
sdo
2649
0.8
2171
1896
54
0.6
sdp
3003
0.6
2509
1690
141
1.0
n
1249
1249
223
55
37
37
A lower limit of simulated nitrate loads is visible in each of the graphs. This limit can
be explained by the fact that only monthly average data on point source effluents were
available for model input, which leads to an overestimation of nitrate loads during low
flows in the model. As illustrated in Fig. 4.6, SWAT-N accurately predicts low and average
nitrate loads for the gauge Obere Dill. However, the model overpredicts high nitrate loads
on a daily basis.
The unsatisfactory model efficiency for the prediction of daily data can be partly explained by input data uncertainty. Only long-term averaged input data of point-source
effluents were available for this study, but we attempted to predict nitrate loads on a daily
basis. Furthermore, assumptions regarding crop rotations as well as timing and amount
of fertilisation have to be made (Lenhart, 2003).
Apart from data uncertainty, the model performance in springtime points to a structural
problem as well. The SWAT-N model hypothesises that frost kills the entire microbes
population, which immobilises nitrogen as well as carbon during the microbial growth. As
soon as temperatures rise above 0 C, the organic nitrogen in the form of dead microbes is
rapidly mineralised in the model. The impact of freeze-thaw events on rapid decomposition
is known in soil science, though there is a large uncertainty in the amount of nitrificatory as
well as denitrificatory nitrogen losses, and the contribution to mineral nitrogen compounds
(Ludwig et al., 2004). The large mineral nitrogen supply in the model due to freeze-thaw
events during spring leads to high amounts of leachable nitrate. The simulated wash-out
of nitrate in spring is visible in both graphs (Figs. 4.5 top and 4.6), though the observed
spring peak is less.
The efficiency of SWAT-N predictions found in this study is similar as reported in other
eco-hydrolocical simulation experiments in mesoscale catchments. Wade et al. (2002) used
the INCA model to predict weekly nitrate concentrations at the river Kennet (1200 km 2 )
in the UK for various sampling sites along the main channel, and found Nash & Sutcliffe
efficiencies between < 0 to 0.8. Jarvie et al. (2002) found E’s between < 0 to 0.5 when
INCA was used for the simulation of nitrate concentrations in the river Tweed (4400 km 2 ),
UK. The model LASCAM predicted monthly loads for the Avon River (119000 km 2 ) and
74
1.2 × 103
6 × 102
Simulated
Observed
0
1.1.00
1.7.00
1.1.01
1.7.01
1.1.02
1.7.02
1.1.03
Date
Figure 4.6: Simulated and observed daily nitrate load at gauge Obere Dill (1.1.2000 28.2.2003).
Ellen Brook (700 km2 ), Australia, with efficiencies of E = 0.61 and 0.89, respectively
(Viney et al., 2000).
Other authors working with the previous SWAT-EPIC version, partly found lower agreements between simulated and observed nitrate loads as compared to this study. Grizzetti
et al. (2003) found an E = 0.30 when they used SWAT at the Vataanjoki watershed
(1680 km2 ), Finland, to model diffuse emissions and retentions of nutrients on a monthly
basis. Chaplot et al. (2004) found R2 = 0.73 for the prediction of monthly nitrate loads in
the Walnut Creek watershed (51 km2 ), Iowa, and Santhi et al. (2001a) reported a model
efficiency (E = 0.64) for monthly nitrate loads at the Bosque River watershed (4277 km 2 ).
In comparison to a previously conducted simulation experiment with the SWAT-EPIC
model in the Dill catchment by Lenhart et al. (2003), who achieved a model efficiency for
monthly nitrate load of E = 0.31, the modifications implemented in SWAT-N improved
the model efficiency. Lenhart et al. (2003) used annual estimated point-source effluents
derived from population equivalents to calculate a point-source background concentration
of nitrate by division with simulated discharge. This background concentration was then
subtracted from observed nitrate concentrations at gauge Aßlar, and multiplied with simulated discharge to yield nitrate loads for reference. Hence, Lenhart et al. (2003) did not
account for any simulations error regarding discharge predictions, whereas both discharge
prediction as well as nitrate load prediction reflecting monthly point-source inputs were
evaluated in this study.
4.3.3 Longitudinal profile of nitrate concentrations
Figure 4.7 illustrates observed and simulated nitrate concentrations along the main channel Dill as well as its tributaries at four different dates. The sampling sites are given in
75
Fig. 4.2 b. Two sampling campaigns were conducted during low flow phases on 10.9.2003
and 15.9.2004. The white columns of the top graph indicate that observed nitrate concentrations are lowest at the Obere Dill (sites 7, 12, 18), strongly increase at site 28, which
is located below the Dietzhölze tributary, and decrease along the main channel until the
gauge Aßlar (site 57). The high nitrate concentrations at site 28 can be explained by the
nitrate rich wastewater released by the steel mill, which is located at the mouth of the
Dietzhölze.
The sampling campaign on 22.1.2004 and the campaign on 13.5.2004 are referred to high
flow and intermediate flow conditions, respectively. Although the nitrate profile peaks at
site 28 in May 2004 during intermediate flow conditions, the clear impact of point source
effluents on nitrate concentrations along the longitudinal profile disappears due to the
higher dilution capacity of the stream during intermediate and high flow periods.
SWAT-N simulates the longitudinal profiles for the dates 22.1.2004 (D = 0.74, R 2 =
0.57), 13.5.2004 (D = 0.70, R2 = 0.30), and 15.9.2004 (D = 0.90, R2 = 0.66) accurately.
However, SWAT-N over-predicts nitrate concentration for site 18 on 15.9.2004, which
can be attributed to incorrect input data regarding point-source effluents, because only
long-term average monthly data were available for the municipal sewage treatment plant
located at this site.
Although SWAT-N performs well at the main channel Dill, the predictions at the tributaries are less accurate (e.g. 15.9.2004). Furthermore, SWAT-N fails to predict the longitudinal profile on 10.9.2003 (D = 0.52, R2 = 0.64). This is because SWAT-N does not
accurately simulate both discharge and nitrate emissions for the days when the longitudinal profile was sampled, although the overall model efficiency in regard to time series
predictions is acceptable. It should be noted that the sampling campaign on 10.9.2003
was conducted after an exceptionally long dry period. Higher discharge predictions as
compared to observed discharge under low flow conditions led to a higher dilution capacity in the model for point source effluents and, hence, lower nitrate concentrations than
observed. The opposite is true for lower flow predictions as compared to observations. It is
noteworthy that SWAT-N simulates nitrate concentrations in the same range as observed
on 10.9.2003 with a time lag of about 10 days (data not presented).
4.3.4 Geographical sources of nitrate leaching
Fig. 4.8 a shows that simulated average annual nitrate leaching through lateral flow that
enters the stream is low for large areas in the Dill catchment (0 to 10 kg N ha −1 ) for the
period 2000 to 2004. At river valleys, in the north of the Dietzhölze catchment, and in
the east of the Aar catchment patches with nitrate leaching between 15 to 40 kg N ha −1
are present according to the model. These patches correspond to pasture land as well as
arable land, where fertilisation takes place.
Figure 4.8 b illustrates the proportion of predicted lateral nitrate transport to predicted
total nitrate transport to the stream. A zone with proportions of up to 90% is present
in the centre of the catchment. In the north, the proportion accounts for up to 50%, and
in the west of the catchments proportions of below 50% are simulated. As previously
shown in Fig. 4.4, SWAT-N simulates highest groundwater contributions to the streams
in the west of catchment, and, thus, lowest lateral flow in this area. Consequently, lateral
76
28
34
38
43
50
53
56
39
28
34
38
43
50
53
56
10
0
57
27
33
35
36
39
13 May 2004, intermediate flow
13 May 2004
34
38
43
50
53
56
10
57
27
33
35
36
39
15 Sep 2004
28
34
38
Dill
43
50
53
56
57
51
0
10
20
Tributaries
NO3 (mg N l−1)
Dill
20
18
0
20
28
51
20
Tributaries
NO3 (mg N l−1)
Dill
18
51
20
22 Jan 2004
10
12
36
22 Jan 2004, high flow
0
7
35
Tributaries
10
12
33
Dill
0
7
20
27
NO3 (mg N l−1)
18
10
57
20
12
0
NO3 (mg N l−1)
20
10
18
10
7
NO3 (mg N l−1)
12
0
NO3 (mg N l−1)
7
NO3 (mg N l−1)
10 Sep 2003
0
NO3 (mg N l−1)
27
33
35
36
39
51
Tributaries
Figure 4.7: Comparison of observed (white) and simulated (red) nitrate concentrations at various sampling locations along the river channel Dill (left) and at various tributaries
(right) on four different dates. The sampling sites are given in Fig. 4.2 b.
77
(kg N / ha)
b)
80
Dietzhölze
20
10
(%)
a)
30
Dietzhölze
60
40
Obere Dill
Obere Dill
20
0
Aar
Aar
Dill
0
Units (m) 10000
Dill
0
Units (m) 10000
Figure 4.8: Simulated average annual nitrate leaching through lateral flow (kg N ha −1 ) for
2000 to 2004 (left). Ratio between nitrate leaching through lateral flow and total
nitrate leaching (%) for the same period (right).
nitrate transport is less in the west. The high proportion of laterally transported nitrate
predicted by the model is consistent with the high lateral flow, which is the dominant
run-off component in mountainous areas.
4.4 Conclusions
Performance regarding to daily discharge predictions at various sites. The model
efficiency for daily discharge predictions at the catchments outlet is on an acceptable level.
The selected time-period and, hence, the selected weather period for calibration has a
significant impact on the model performance but it does not significantly influence the
’optimal’ parametrisation found by SCE-UA.
It can be concluded from the proxy-basin test that an ’optimised’ parameter set for the
entire catchment will lead to accurate predictions of daily discharge at up-stream gauges,
if the geology and, hence, hydrologically effective parameters of the specific subbasin do
not significantly differ from the entire catchment, as is the case for the Obere Dill.
Performance regarding daily, weekly, and monthly nitrate load predictions
and geographical sources of nitrate emissions.
The modifications incorporated in SWAT-N improved the simulation of the nitrogen
cycle as demonstrated by Pohlert et al. (2005c) and as compared to a previous study with
the SWAT-EPIC version by Lenhart et al. (2003).
A visual inspection of simulated nitrate load and observations at gauge Aßlar suggests
a reasonable performance of the model for daily nitrate load predictions. However, the
78
the model efficiency (E = −0.29) for daily data indicates that the average of observations
is a ’better’ predictor than the SWAT-N model!
The values of goodness-of-fit measures will increase to an acceptable degree, if the
nitrate load predictions are averaged to a weekly or a monthly basis for the gauge Aßlar.
Consequently, we conclude that SWAT-N can be used for monthly, as well as weekly
predictions of nitrate load, but should be avoided for daily predictions.
The comparison of sampled longitudinal profiles of nitrate with model predictions suggests that simulated nitrate emissions from both non-point and point sources are spatially
consistent. This new approach to test the internal model performance with one-point-intime many-points-in-space was a useful addition to the ’classical’ split sample techniques,
where one-point-in-space and many-points-in-time are considered.
Can SWAT-N be used for scenario assessment? Based on an evaluation of the
performance of any eco-hydrologic model, the modeller has to decide whether to reject
the model or to apply the model for a specific purpose.
We conclude that SWAT-N can be used to predict changes on discharge caused by landuse changes on a daily basis. The accuracy of nitrate load predictions is acceptable when
SWAT-N is applied on a monthly or weekly basis. The model can be used for scenario
assessment, because it uses an enhanced process-description and it is a robust approach
that needs no further calibration. It provides a promising tool for the joint research within
the SFB 299 as well as the assessment of action plans for the implementation of the EUWater Framework Directive (2000/60/EG, 2000). However, the use of the model should
be avoided for assessing land-use impacts on nitrate loads on a daily basis.
The research project is granted by the Deutsche Forschungsgemeinschaft in the framework
of the Sonderforschungsbereich 299 at the Justus-Liebig-University Gießen. We would like
to acknowledge the help of Beate Lindenstruth, Nelly Weis, and Adrian Skitschak who
helped to maintain the sampling campaign conducted at the Dill River.
79
80
In den folgenden Anhängen werden die Funktionsweise verschiedener Prä- und Postprocessing Werkzeuge für SWAT und SWAT-N vorgestellt, sowie die Konfiguration von SWAT-N
Simulationsläufen erläutert. Bei sämtlichen Programmen handelt es sich um FORTRAN
Programme, deren Quellcode dem Institut für Landschaftsökologie und Ressourcenmanagement zur Verfügung stehen und als Konsolenprogramme unter DOS oder LINUX laufen.
Die Tabelle A.1 gibt eine Übersicht über die entwickelten Programme und Empfehlungen
für deren Anwendung. In der Spalte Verwendung wird zwischen optional und obligatorisch
unterschieden. So ist zum Beispiel die Verwendung von addph obligatorisch, sofern mit
SWAT-N gearbeitet werden soll. In der Spalte Status werden drei Ebenen unterschieden:
Stabil Der Quellcode wurde auf verschiedenen Plattformen erfolgreich getestet und auf
Anwendungsfreundlichkeit programmiert. Das Programm ist ohne Programmierkenntnisse anwendbar.
Test Der Quellcode ist getestet worden, jedoch nicht auf Anwendungsfreundlichkeit
programmiert. Programmierkenntnisse sind bei der Fehlersuche (Eingabedaten)
nützlich, jedoch nicht zwingend erforderlich
Entwicklung Der Quellcode wurde zwar getestet, jedoch sind für die Anwendung Programmierkenntnisse in FORTRAN erforderlich. Fehler im Quellcode sind möglich.
Das Programm READOUT wurde vom Autor unter den Bedingungen der General
Public License veröffentlicht. Es kann von der SWAT-Homepage (www.brc.tamus.edu/
swat/) heruntergeladen werden.
Die anderen Programme sind intellektuelles Eigentum des Urhebers und für den Eigengebrauch des Instituts entwickelt worden. In der derzeitigen Form sollten diese
Programme nicht an Dritte abgegeben werden. Die Programme sind in dieser Form
nicht auf Anwendungsfreundlichkeit hin programmiert worden. Ferner fehlen rechtliche
Tabelle A.1: Hilfsprogramme für die Modellierung mit SWAT-N
Funktion
Verwendung Status
Simulationsprogramm zur WasserTest
und Stickstoffhaushaltssimulation
addph
Hinzufügen von pH-Werten in die .sol Dateien
Obligatorisch Stabil
idw
Interpolation von Niederschlagswerten (pcp.pcp) Optional
Entwicklung
READOUT Auslesen der .rch, .bsb oder .sbs Dateien
Optional
Stabil
GoF
Berechnet verschiedene Gütemaße
Optional
Stabil
hru2map
Erstellt thematische Karten aus .sbs Datei
Optional
Test
Programm
swatg n
81
Hinweise wie Haftungsausschluss des Autors und des Instituts. Dem Institut für Landschaftsökologie und Ressourcenmangement ist es freigestellt, die Programme dahingehend
weiterzuentwickeln und vorbehaltlich des Haftungsausschlusses an Dritte herauszugeben.
A.1 A short manual for SWAT-N
A.1.1 Introduction
This is a brief manual for the use of SWAT-N, which includes algorithms for the simulation
of N-cycle and N-transport based on the DNDC-model (Li et al., 1992, 2000; Zhang et al.,
2002), the CropSyst model (Stöckle et al., 2003) and the hydrologic model SWAT-G
(Eckhardt et al., 2002). To prevent any confusions regarding the versions of SWAT, a
brief history of the development is given.
The model SWAT-G bases on the code of SWAT-99.2. Eckhardt et al. (2002) included
some algorithms to account for the anisotropy of hydraulic conductivity reagarding vertical
and lateral flow. This resulted in an increase of interflow, and an improved accuracy for
discharge prediction for the simulated Dill catchment.
Since 2003 the ILR changed to the SWAT2000 version (Neitsch et al., 2002) and J.
A. Huisman implemented those modifications made by Eckhardt et al. (2002) into the
SWAT2000 code. This SWAT version may be referred to as SWAT2000-G; however, in
publications we still use the term SWAT-G.
Pohlert included the aforementioned algorithms for N-cycling into the code of
SWAT2000-G, which ends up in the version SWAT-N.
A.1.2 How the program works
Any user of SWAT-N has the choice to run the model with the original approach which
bases on EPIC (see Neitsch et al., 2002, for details) or the newly implemented approach
which bases on DNDC and CropSyst.
Configuration of the run
The executable is a two-in-one program. To run the SWAT-N version, a file named
basins_dndc.con must exist in the project directory (txtinout). If this file does not
exist, the SWAT-G version (SWAT2000 with Eckhardt’s modifications) with the EPIC
approach is executed.
The content of the file specifies the catchment wide parameters that govern the Ncycle. The first line of the file is a comment header which is ignored by the program. The
paramers are read in with free-format, however, the sequential order of the parameters
does matter, i.e. the program assumes that the value for krcvl is always in line 2. This
file is also taken into account by SCE-UA, with which an automatic calibration can be
conducted.
!! This file specifies parameters for SWAT-DNDC.
2.500E-12 ! 1 krcvl
82
A.1 A short manual for SWAT-N
7.400E-12
2.000E-12
025.000000000
000.010000000
000.330000000
000.040000000
000.900000000
000.425000000
000.150000000
000.160000000
000.200000000
000.670000000
000.200000000
000.025000000
045.000000000
033.300000000
001.900000000
058.000000000
034.150000000
005.390000000
010.000000000
005.000000000
050.000000000
200.000000000
010.000000000
001.500000000
000.001350000
000.500000000
2
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
2 krcl
3 krcr
4 kni35
5 kdni15
6 KRB
7 hrb
8 SRB
9 w_crit
10 w_delta
11 srh
12 EFFNO
13 EFFRB
14 EFFAC
15 D_nh3
16 tmax
17 topt
18 ta
19 tmax2
20 topt2
21 ta2
22 rcnb
23 rcnrvl
24 rcnrl
25 rcnrr
26 rcnh
27 max_nup
28 DDRF
29 anr
30 TMPSMF
Configuration of SWAT-N output
The output is configured with two files, namely basins.cod and dndc1.out. With
basins.cod the output for the reach file (basins.rch) and basins file (basins.bsb)
can be controlled.
It is worth to note, that N-relevant output variables simulated with SWAT-N are not
correctly written into the basins.sbs file, where HRU related output is usually written into!
A new writing routine that writes N related daily output for each HRU was implemented in the code to overcome the aforementioned problem. In case of a non-existing
dndc1.out, some 22 variables for each day and HRU will be written into an output file
named dndc2.out. On the other hand, a user may select some specific output variables
by editing the control file dndc1.out. It is noteworthy, that SWAT-N always starts printing the value of a selected variable that is simulated for the first day of simulation. An
annual skip of output, as it is possible with other SWAT-G output variables, and as it is
controlled by basins.cod, is not possible with HRU-related output. The content of this
83
file may look like this:
1
2
15
In this case, three variables will be written into the output file dndc2.out. The below
list provides the meaning of each identifier:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Daily mineral nitrogen content for the profile (kg N /ha)
Total organic nitrogen for the profile (kg N / ha)
Ammonia (kg N /ha)
Ammonium (kg N /ha)
Nitrate (kg N / ha)
Absorbed N on clay particles (kg N / ha)
Organic N in very labile pool (kg N /ha)
Organic N in labile pool (kg N /ha)
Organic N in stabile pool (kg N /ha)
Crop N uptake (kg N / (ha d))
(NOTE: A bug)
N in crop yield (kg N / (ha d))
(NOTE: A bug)
N fertilised in HRU (kg N / (ha d)) (NOTE: A bug)
N in precipitation (kg N /(ha d))
Total N leaching (water transport) (kg N / (ha d))
N leaching due to surface runoff (kg N /(ha d))
N leaching due to lateral flow (kg N /(ha d))
N leaching due to percolation (kg N / (ha d))
Total N loss into the atmosphere (kg N / (ha d))
N emission due to denitrification (kg N / (ha d))
N emission as NO (kg N /(ha d))
N emission as N2O (kg N / (ha d))
N emission due to volatilisation (kg N / (ha d))
A.1.3 Pre-processing
SWAT-N bases on the hydrological concept of SWAT-G. That is why a correct SWATG project needs to be set up first, i.e. values for anisotropy of hydraulic conductivity
must be specified in the .sol files in line 19. The program parset.exe was written by
Eckhardt and modified by Huisman to do this job. Furthermore, the concept of N-cycle
within SWAT-N includes the impact of soil pH. This is why each .sol file must contain
an additional line for pH-values (line 20). The batch-program addph was written to do
this job. However, parset can be used for this as well, but addph is more flexible. The
steps for pre-processing are briefly given below:
84
Set up a project with the AVSWAT interface as given by Di Luzio et al. (2001).
Check, that SWAT2000 runs without any errors. Verify, that land use management
(i.e.: crop rotation, fertilisation) is simulated correctly. Otherwise, check the mangement files (.mgt).
A.2 addph
Run Addnumber.exe (see Huisman’s manual “Preprocessing of projects outside the
Arcview-Interface”).
In mountaineous catchments, where shallow soils above almost impermeable rocks
are present, an included bedrock layer might improve SWAT-predictions with regard
to simulated interflow. Either the AVSWAT interface can be used to include such
an additional bedrock layer for given soil-types or the batch program Addhor.exe.
The latter program writes parameters for a bedrock layer in each .sol file (see
Huisman’s manual). The input-file is horizon.txt.
The anisotropy-factor can now be written by parset.exe (see Huisman’s manual).
Run addph.exe, which will add a new line into each .sol file containing pH-values
for each horizon. The input-file is pH.txt. This program is explained in a different
manual.
A.1.4 Post-processing
A tool hru2map was written, which re-locates the HRU specific output for a single day to
a grid map. Hence, the results of basins.sbs as well as dndc2.out can be visualised or
further manipulated using GIS.
A.1.5 Note for programmers
Table A.2 provides information regarding the modified a newly created source files.
A.2 addph
A.2.1 Einleitung
Um die erweiterte Version SWAT-N nutzen zu können, werden pH-Werte des Bodens
benötigt. In SWAT-N steuern pH-Werte sowohl das Ammonium - Ammoniak Verhältnis
(siehe dazu dndcclayadsorb.f90), als auch die Kinetik der Nitrifikation (Ammonium
zu Nitrat) (siehe dndcnitri.f90). Letzteres ist abhängig vom Gehalt des organischen
Stickstoffs, der in der Regel nur im Oberboden (1. Horizont) vorkommt (vgl. dazu CORG Gehalte der Horizonte in Bodenprofil-Ansprachen).
A.2.2 Was ist das Problem?
Das Preprocessing-Werkzeug AVSWAT (Di Luzio et al., 2001) erzeugt Eingabedateien
für die Programmversion SWAT2000. Da in dieser SWAT-Version keine pH-Werte als
steuernde Größen vorgesehen sind, schreibt AVSWAT folgerichtig keine pH-Werte in die
Bodendateien (.sol).
Um ein mühsames Eintragen der pH-Werte in die zahlreichen .sol Dateien von Hand
zu vermeiden, wurde das Programm AddpH geschrieben.
85
Table A.2: New and modified source files included in SWAT-G-N
allocate parms.f
dndcallocate.f90
dndcalloclit.f90
dndcclayadsorb.f90
dndcdenit.f90
dndcdormant.f90
dndcfert.f90
dndcgraze.f90
dndcharvestop.f90
dndcharvkillop.f90
dndckillop.f90
dndcnitri.f90
dndcnminrl.f90
dndcnrain.f90
dndcnup.f90
dndcorgn.f90
dndcparam.f90
dndcsoil chem.f90
dndcsum.f90
dndcvolatil.f90
grow.f
harvkillop.f
main.f
nfix.f
nlch.f
nminrl.f
nup.f
openfile.f
operatn.f
readbsn.f
readcod.f
read dndc.f90
readinpt.f
readsol.f
rtover.f
sbsmon.f
soil phys.f
solt.f
subbasin.f
tillmix.f
zero0.f
A.2.3 Was kann das Programm?
AddpH kennt insgesamt 4 Optionen, wie pH-Werte den Bodendateien zugewiesen werden
können:
1. Nach Bodentyp (Soil-ID) getrennt, gleicher pH-Wert pro Horizont.
2. Nach Landnutzungsklasse (LU-ID) getrent, gleicher pH-Wert pro Bodenhorizont.
3. Nach Hydro Agri Dülmen (1993, S. 265-266), sofern keine relevanten Informationen
über pH-Werte im EZG vorliegen.
4. Nach Bodentyp (Soil-ID) und Horizont, sofern vollständige Profilangaben vorhanden
sind.
Für die Arbeiten im EZG der Dill wurde die 2. Variante gewählt, da im verfügbaren
Bestand der Bodenflächendaten 1:50000 (BFD50) Angaben zu pH-Werten, bzw. pH-Wert
Stufen (noch) fehlten. Die zugewiesenen pH-Werte wurden nach Landnutzungsklassen aus
der Datenbank des SFB-Projekts von Prof. Gäth entnommen. Für die Arbeiten in den
EZG Vils und Rott in Bayern, lagen Leitprofil-Daten für vektorisierte BÜK25 Karten vor,
sodass die Variante 4 gewählt wurde.
Das Hessische Landesamt für Umwelt und Geologie bietet landesweit die BFD50 für
Forschungszwecke an und aktualisiert die dazugehörigen Datenbanken fortlaufend. Je nach
Aktualisierungsstand können aus den dazugehörigen Datenbanken pH-Werte entnommen
86
A.2 addph
werden (Variante 4). Die pH-Angaben liegen bei BFD50 als pH-Stufen vor, die nach
AG Boden (1994) in den negativ-dekadischen Logarithmus (pH-Wert) überführt werden
können.
Bei der Variante 3 können die pH-Werte als Ziel-pH-Werte nach bester landwirtschaftlicher und gartenbaulicher Praxis (BLP) (bezieht sich auf Ertragsoptimierung, nicht Umweltschonung!) aufgefasst werden und stellen eine Notlösung dar. Diese Variante bietet
aber auch die Möglichkeit, im ITE2 M Modellverbund des SFB299, pH-Werte nach den von
ProLand gelieferten potentiellen Landnutzungskarten zuzuweisen (Annahme von BLP in
ProLand).
A.2.4 Benutzung von AddpH
Das Programm fügt in die 20. Zeile jeder .sol Datei einen pH-Wert Eintrag hinzu. Hierzu
ist unbedingt folgendes zu beachten:
Das Projekt muss vollständig und fehlerfrei mit AVSWAT aufgebaut worden sein.
(Prüfe, ob ein “normaler” SWAT2000 Lauf fehlerfrei funktioniert!)
Das von S. Huisman beschriebene Prozedere der Vorbereitung eines SWAT-G Projektes außerhalb der AVSWAT Umgebung muss vor der Benutzung von AddpH stattgefunden haben!
Eine Sicherungskopie für sämtliche .sol Dateien wird empfohlen.
Eine für die Bearbeitung mit AddpH vollständige .sol Datei sieht z. B. wie folgt aus:
.sol file Subbasin:6 HRU:7 Luse:FRSD Soil:Soil-2458 Wed Feb 18
Soil Name: Soil- 2458
3: 2458
Soil Hydrologic Group: D
Maximum rooting depth(m) : 1100.0
Porosity fraction from which anions are excluded: 0.500
Crack volume potential of soil: 0.500
Texture 1
: Ut3
Ut3
U
Depth (mm)
:
50.000
500.000
1100.000
Bulk Density Moist (g/cc):
1.355
1.557
1.557
Ave. AW Incl. Rock Frag :
0.295
0.206
0.103
Ksat. (est.) (mm/hr)
:
199.136
71.120
71.120
Organic Carbon (weight %):
3.500
0.290
0.000
Clay
(weight %):
14.150
14.150
14.150
Silt
(weight %):
74.630
74.630
74.630
Sand
(weight %):
11.220
11.220
11.220
Rock Fragments
(vol. %):
17.500
17.500
62.500
Soil Albedo (Moist)
:
0.220
0.250
0.300
Erosion K
:
0.600
0.600
0.600
Anisotropy factor
:
0.000
1.149
2.298
87
Ferner wird eine Eingabedatei Namens ph.txt benötigt. Der Aufbau dieser Datei ist in
den weiteren Kapiteln beschrieben. Grundsätzlich kann für die Erstellung einer ph.txt ein
Tabellenkalkulationsprogramm wie EXCEL verwendet werden, sofern die Tabelle dann in
das ASCII-Format ausgeschrieben wird (siehe hierzu auch Kapitel A.2.6).
Variante 1
Bei dieser Variante werden nach Bodeneinheiten getrennt, gleiche pH-Werte pro Horizont zugewiesen (vgl. S. 86). Auszugsweise ist eine solche ph.txt mit Kommentaren 1
dargestellt:
1
100 4.5
101 5.5
103 5.5
2223 7.5
...
!! 1. Zeile: Schalter für die gewählte Variante
!! 2. Zeile bis Ende der Datei
!! SOIL-ID pH-WERT
Das Trennzeichen ist ein Leerzeichen. Achtung: Die maximale Zahl für SOIL-ID ist 9999.
Variante 2
Bei dieser Variante werden nach Landnutzungsklassen einheitliche pH-Werte für jeden Horizont zugewiesen. Das Programm unterscheidet nicht zwischen Groß- oder Kleinschreibung, jedoch muss die Zeichenkette für die Landnutzungsklasse aus exakt 4 Buchstaben bestehen (vgl. hierzu auch Di Luzio et al., 2001, S. 13). Für die Kategorie Wasserfläche (WATR) gibt man am Besten den Wert -99.0 an. SWAT-N berechnet keine
N-Mineralisierung auf Wasserflächen.
2
AGRL
FRSD
FRSE
URBN
PAST
WATR
5.9
4.1
4.1
7.0
5.1
-99.0
!! LANDUSE pH-WERT
Variante 3
Für die Variante 3 ist der Aufbau der ph.txt eine Tabelle mit 3 Feldern. Das erstes
Feld definiert die Landnutzungsklasse, das zweite Feld die Tonklasse (vgl. Hydro Agri
Dülmen, 1993, S. 265-266) und das dritte Feld den pH-Wert. Da bei Hydro Agri Dülmen
(1993) nur Angaben für Grünland und Acker gemacht werden, sind pH-Werte für andere
Landnutzungsklassen aus anderen Quellen zu entnehmen. Für Wälder, Gewässer und
Siedlungsflächen muss im zweiten Feld jeweils -99 angegeben werden.
1
Für den Programmbetrieb, keine Kommentare verwenden!
88
A.2 addph
3
AGRL
AGRL
AGRL
FRSD
FRSE
URBN
PAST
PAST
PAST
WATR
!! Werte aus Agri Hydro Dülmen (1993)
1 5.8
2 6.6
3 6.8
-99 4.5
-99 4.3
-99 7.0
1 4.9
2 5.4
3 5.8
-99 -99.0
Variante 4
Für diese Variante werden je nach Bodentyp und Horizont, pH-Werte zugewiesen. Das
erste Feld enthält die SOIL-ID, das zweite Feld enthält die Anzahl der Horizonte und die
Felder 3 bis n enthalten die pH-Werte für die Horizonte 1 bis n. Der Auflagehorizont (Ap
oder Ah) ist hierbei immer der Erstgenannte.
4
100 4 4.5 4.2 6.5
101 5 5.5 5.1 4.5
!!
103 2 4.2 3.2
2223 6 5.4 5.4 5.0 4.8 4.2 4.1
...
A.2.5 Ergebnis
Benutzt man die Eingabedatei ph.txt aus Kapitel A.2.4, dann ist das Ergebnis für alle
.sol Dateien, in der die Landnutzungsklasse FRSD vorkommt, wie folgt:
.sol file Subbasin:6 HRU:7 Luse:FRSD Soil:Soil-2458 Wed Feb 18
Soil Name: Soil- 2458
3: 2458
Soil Hydrologic Group: D
Maximum rooting depth(m) : 1100.0
Porosity fraction from which anions are excluded: 0.500
Crack volume potential of soil: 0.500
Texture 1
: Ut3
Ut3
U
Depth (mm)
:
50.000
500.000
1100.000
Bulk Density Moist (g/cc):
1.355
1.557
1.557
Ave. AW Incl. Rock Frag :
0.295
0.206
0.103
Ksat. (est.) (mm/hr)
:
199.136
71.120
71.120
Organic Carbon (weight %):
3.500
0.290
0.000
Clay
(weight %):
14.150
14.150
14.150
Silt
(weight %):
74.630
74.630
74.630
Sand
(weight %):
11.220
11.220
11.220
89
Rock Fragments
(vol. %):
Soil Albedo (Moist)
:
Erosion K
:
Anisotropy factor
:
pH value
:
17.500
0.220
0.600
0.000
4.100
17.500
0.250
0.600
1.149
4.100
62.500
0.300
0.600
2.298
4.100
A.2.6 Limitierungen
Länge der Zeichenkette für Landnutzungsklasse ist 4 Zeichen.
Maximale Zahl für die SOIL-ID ist 9999.
Trennzeichen ist ein Leerzeichen (keinesfalls TABSTOPP).
Dezimaltrennzeichen ist Punkt.
Maximale Anzahl der Horizonte ist 10.
A.3 idw – Inverse Distant Weighted Average
Das Programm idw hat zum Zeitpunkt des Verfassens dieses Textes den Status Entwicklung, d. h. FORTRAN-Kenntnisse sind für eine erfolgreiche Benutzung erforderlich. Die
Tabelle A.3 listet die relevanten Quellcode-Dateien des Programmpakets auf.
Tabelle A.3: Liste der Programmdateien zu idw.
Datei
Funktion
main.f90
Hauptprogramm mit idw-Verfahren
readcentroids.f90 Liest die Koordinaten der Zentroide
readpcpdat.f90
Liest die Niederschlagswerte der Stationen
readstatloc.f90
Liest die Koordinaten der Stationen
countnil.f90
Zählt fehlende Werte
A.3.1 Ziel
Ziel des Programms idw ist die räumliche Interpolation der täglichen Niederschlagswerte
für jeweils ein Teileinzugsgebiet (subbasin) basierend auf den Werten der umgebenden
Niederschlagsstationen. Somit kann mit dem Programm idw für jedes Teileinzugsgebiet
jeweils eine Niederschlagsdatei erstellt werden. Ein wesentlicher Vorteil dieses Verfahrens
gegenüber dem Nearest-Neighbour-Verfahren in AVSWAT ist die Tatsache, dass auch Stationsreihen mit fehlenden Werten verarbeitet werden können. Das Programm interpoliert
nach der in Fürst (2004) gegebenen Beziehung, wobei:
PN
i=1
ẑ (x0 , y0 ) = PN
90
z(xi ,yi )
hβ
i,0
1
i=1 hβ
i,0
hi,0
q
= d2i,0 + δ 2
(A.1)
A.3 idw – Inverse Distant Weighted Average
mit ẑ (x0 , y0 ) interpolierter Wert für die Koordinate (x0 , y0 ) des Flächenzentroiden eines Teileinzugsgebiets, N Anzahl der Stützstellen, di,0 Distanz zwischen Zentroid und
Niederschlagsstation, δ Glättungsparameter und z (xi , yi ) beobachteter Wert an der Niederschlagsstation und Exponent β. Für β = 2 und δ = 0 geht Gl. A.1 in die übliche Form
über, d.h. das Quadrat der Distanz wird für die Interpolation berücksichtigt.
A.3.2 Funktionsweise
Das Programm benötigt eine Stationstabelle mit Koordinaten, eine Datei mit den Stationsniederschlägen, sowie eine Datei mit den Koordinaten der Flächenzentroiden. Für
jeden Tag und jeden Flächenzentroid berechnet idw einen Niederschlagswert. Für die Interpolation werden die Stationen berücksichtigt, die sich innerhalb eines vom Benutzer/In
interaktiv anzugebenen Distanz befinden (lokale Interpolation). Die Dateiformate der Eingangsdateien ist den entsprechenden Fortran-Routinen zu entnehmen (vgl. Tabelle A.3).
Als Ergebnis liefert idw für jedes Teileinzugsgebiet jeweils eine ASCII-formatierte Datei
mit interpolierten Niederschlagsdaten. Diese Dateien können nun als “Niederschlagsstationen” wiederum in AVSWAT eingelesen und dann mit AVSWAT normal weiter verarbeitet
werden. Da sich nun die Niederschlagsdaten auf die Koordinaten der Zentroide der Teileinzugsgebiete beziehen, wird dann beim Nearest-Neighbour-Verfahren exakt diese Datei
mit den interpolierten Niederschlägen ausgewählt.
A.3.3 Wichtige Hinweise
Zur Berechnung der Koordinaten der Zentroide kann die als Public Domain
erhältliche Extension für ArcView GIS “Add XY extension” von Z. L. Stauber
verwendet werden (siehe http://arcscripts.esri.com/). Diese fügt der Attributtabelle (.dbf Datei) der Teileinzugsgebietskarte im AVSWAT-Projekt jeweils X
und Y-Koordinaten für den Flächenzentroiden hinzu.
Es ist ratsam, die Einleseformate für die Eingabedateien in den Routinen zu ändern.
Es ist unbedingt ratsam, die Datei readpcpdat.f90 zu ändern. Diese korrigiert zur
Zeit Tagesniederschlagswerte von P ≤ 0.2 auf P = 0, da dies für ein anderes Projekt
notwendig war.
In der main.f90 müssen jeweils die Anzahl der Zentroide (censize), die Anzahl
der Stationen (statmx) und die Anzahl der Messwerte pro Station (recl) manuell
geäandert werden und entsprechend neu kompiliert werden.
ACHTUNG: Ist die maximale Distanz zu klein gewählt, dann werden keine Stationswerte innerhalb des angegeben Radius gefunden. Das führt dazu, dass Gleichung
A.1, S. 90 keine Funktionswerte liefert. In den Ergebnisdateien steht denn die Fehlkennung NaN.
91
A.4 READOUT
A.4.1 What is READOUT?
READOUT is a menu driven, interactive program to extract a variable of interest from
the major SWAT2000 output files, namely basins.rch, basins.sbs and basins.bsb. You can
extract one variable per subbasin/reach or HRU.
Since SWAT2000 output files may get very big (> 500 MB), SWAT users find it difficult
to post-process their simulation runs. READOUT helps the user, to extract and write the
data of ONE variable of interest to a file. The output file of READOUT can easily be
imported to spreadsheet applications.
Make sure you have the SWAT2000 User’s Manual (Chapter 32, pgs 325-330).
A.4.2 Installation
WIN-2000,NT,XP
READOUT was written in FORTRAN90 and was compiled as a console application for
MS-DOS under WIN2000, WIN-NT or WIN-XP.
1. Download the zip-file ”readout.zip”from the SWAT-homepage (www.brc.tamus.
edu/swat/)
2. Unzip the file. You will get an executable file named READOUT.exe
3. Copy the executable file to the following directory (Let us assume, ”C:”is your
partition, where WindowsNT is installed: C:\WINNT\system32\verb)
4. That’s it.
Installation from source
If you run SWAT2000 on UNIX or LINUX, you may wish to compile READOUT for your
machine. You need ”make”for compilation.
1. Get the tar-file ”readout.src.tar.gz”from the SWAT-homepage (see above).
2. Copy the tar-file in an installation directory (i.e. ”readout”)
mkdir readout
cp readout.tar.gz /MY/PATH/TO/readout/readout.tar.gz
cd /MY/PATH/TO/readout/
3. Unpack the tar-file:
gzip -d readout.tar.gz
tar -xf readout.tar
92
A.4 READOUT
4. Check the ”Makefile”for customization (i.e. change the name and the options of the
compiler).
5. For compilation and linking of READOUT run make:
make
6. If you want to run READOUT from any directory, you have to copy/move the
binary file READOUT to a directory within your shell-path (i.e. /usr/bin) NOTE:
You have to be ”root”to do so:
cp /MY/PATH/TO/readout/READOUT /usr/bin/READOUT
7. Change the file permissions as apropriate (you have to be ”root”):
chown
chgrp
chmod
chmod
root READOUT
root READOUT
ugo-wr READOUT
o+x READOUT
The file permissions would look like that:
ls -l READOUT
-r-xr-xr-x
1 root
root
432900 Mai 10 15:08 READOUT*
A.4.3 Starting READOUT
Since READOUT is a console program, the interface is a command line. Do the following
clickings to open an MS-DOS Window:
|START|PROGRAMS|ACESSOIRES|COMMAND PROMT|
Now it’s time to remember some essential DOS commands:
cd
dir
mkdir
copy
del
CRT-C
exit
Change directory
Directory list
Create Directory (Folder)
Copy a file
Delete a file
Aboard the program
Exit MS-DOS Session
Change with ”cd”to your folder with the SWAT2000 output files.
Start the program with the command ”READOUT”. The READOUT session begins
with a welcome message like this:
93
Welcome to READOUT for SWAT2000
Copyright (C) 2004 Thorsten Pohlert
READOUT is a menu driven program to extract a variable of interest
from the major SWAT2000 output files, namely basins.rch, basins.sbs
and basins.bsb. You can extract one variable per subbasin/reach or
HRU.
(... to be continued!)
You will be guided through a READOUT session.
You can ABORT READOUT without savings at any time by enter CRT-C.
A.4.4 Warning
Annual printout frequency
READOUT does not work, if you have selected ”ANNUAL PRINTOUT FREQUENCY”.
However, the SWAT2000 output files will not get so big with the above option, thus
READOUT is not necessary.
Customized basin.cod
I successfully tested READOUT on customized basin.cod files. Apparently, there was no
problem with customized output.
Customized source code
READOUT will DEFINITELY NOT WORK if you customized the FORMAT specifier
for the WRITE command of any of the following ”writing”routines:
sbs[rch,bsb]aa.f sbs[rch,bsb]day.f sbs[rch,bsb]mon.f sbs[rch,bsb]yr.f
You are free to use and modify the source code of READOUT, to meet your requirements.
A.4.5 Importing READOUT results into other applications
After a successfull READOUT session, you may want to import the results into a spreadsheet application or a graphic program to plot diagrams.
During the session you will be asked for a certain output format. Each output file is
ASCII-Format, but the field delimiter differs. You can chose comma (”,”), semicolon (”;”),
gnuplot format or R format.
Spreadsheets
The best formats for importing the output into a spreadsheet application are commaseparated-fields (”,”) for the English spoken countries and a semicolon (”;”) for non-
94
A.4 READOUT
English spoken countries (since (”,”) is the decimal delimiter in non-English spoken
countries). This may work for importing READOUT files into OPEN-OFFICE, MSEXCEL(R) and SPSS(R).
GNUPLOT
The gnuplot format can be easily evoked by gnuplot. Assume you extracted the
FLOW OUT variable for the reach 8 and wrote the result to the file flowout08.txt, you
will just need to type this command within gnuplot (one line):
> plot ’flowout08.txt’ using 1:3 title \
’Outflow from RCH 8’ with line
This will produce a line diagram with the title ”Outflow from RCH 8”where field 1 (a
counter) and 3 (the variable) are used.
To learn more about the free software package gnuplot, see http://www.gnuplot.info/
R
For plotting and further statistical analysis of your SWAT2000 data you may want to use
the free statistical package R. You can select the R format within READOUT, which enables an easy import into the R environment. We assume again you selected FLOW OUT
from reach 8. To do a line plot, enter the following commands within R.
> xx <- read.table("flowout08.txt")
> plot.default(xx[2], type="line", linetype="solid")
After you plotted your diagram, you can run statistical analysis within R. The command
”summary”will produce an output of descriptive statistics like this:
> summary(xx)
MON
Min.
: 0.00
1st Qu.: 91.25
Median :182.50
Mean
:182.50
3rd Qu.:273.75
Max.
:365.00
X.FLOW.OUTcms
Min.
:0.000e+00
1st Qu.:3.314e-06
Median :9.638e-04
Mean
:2.699e-02
3rd Qu.:1.864e-02
Max.
:6.262e-01
where: MON is the time (here: day of year) and X.FLOW.OUTcms is the discharge leaving
reach 8 of our virtual project.
You can do many more things with R, i.e. time series analysis, statistical testing, . . . To
get more information and a free download version of R, see: http://www.r-project.org/
95
A.4.6 Error messages and limitations
Errors
You will be guided throughout a READOUT session. The error messages are almost
self-explanatory. However, there might be some cases, when READOUT aborts and a
FORTRAN90 error message occurs. The most common one might be the following one.
MSG Error 148: Invalid character
A: You may have typed a ”letter”, instead of a ”number”. Start READOUT again and
do your selections by typing the given digit, not characters.
MSG Error 530 : In program unit [unit] the size of array [array] exceeds the implementation limit (2**31-1)
A I am pretty sure, you will never get this message (see below).
Limits
I carefully created READOUT with the use of the FORTRAN90 feature to dynamically
allocate RAM and the use of a ”file-line-counter”with data type REAL (some .bsb files
may have more lines, then the range of INTEGER data types on a 32-BIT processor).
However, there are some technical limits, although of theoretical importance:
The 2,737-years-limit The output FORMAT for the enumerator is set to 6 digits (max.
Nr. 999999), which equals a SWAT2000 run with daily printout frequency of 2,737
years. If you intend to run SWAT2000 throughout the paleocene, you might want
to change the FORMAT specifier in the writing subroutines :-).
The 2,939,744-years-limit The data will be written into an array, which size may not
exceed the range of the INTEGER data type (2exp31-1 on a 32-BIT processor).
This equals a SWAT2000 run with a daily printout frequency of 2,939,744 years. If
you run READOUT on a 16-BIT processor, the largest INTEGER number will be
less (exactly 2exp16-1). I guess, you will have to recompile READOUT on such a
processor, since it was compiled on a 32-BIT processor.
Memory runs out I successfully tested READOUT on a 750 MB output file. However,
extracting data from such huge files may take several minutes. I realized, that my
computer started to SWAP memory from the RAM to the harddisk. The reason
for this might depend on the operation-system and how it manages huge files, since
READOUT does not use much RAM.
A.4.7 De-installation
For de-installation of READOUT, simply use your file manager (i.e. MS Explorer(R)) and
delete the binary file (READOUT.exe) from your installation folder.
96
A.5 GoF – Goodness-of-Fit
A.4.8 To do
There are some things to do:
Expand the capability of READOUT, that the ”HRU impoundment output
file”(.wtr) and the ”Reservoir Output File”(.rsv) can be extracted.
Expand the capability of READOUT, that averages for each variable and REACH/
SUBBASIN can be computed. This might be useful, for linking the output to GIS.
A.4.9 License (short)
READOUT for SWAT2000 Copyright (C) 2004 Thorsten Pohlert
This program is free software; you can redistribute it and/or modify it under the terms
of the GNU General Public License as published by the Free Software Foundation; either
version 2 of the License, or at your option) any later version.
A.4.10 Disclaimer (short)
This program is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
A.5 GoF – Goodness-of-Fit
A.5.1 Funktionalität
Das Programm berechnet verschiedene Maße zur Bewertung der Modellgüte. Im einzelnen
sind dies:
E Modelleffizienz nach Nash & Sutcliffe (1970). Siehe Gleichung 4.1, Seite 68.
D Wilmott’s index of agreement (Wilmott et al., 1985). Siehe Gleichung 4.2, Seite 68.
RMSE Root mean
P square error: 0.5
RM SE = [ (xo − xs )2 ) /n] .
MAE Mittlerer P
absoluter Fehler:
M AE = ( |xo − xs |) /n.
RRS Root mean residual after sorting. Siehe van Griensven & Bauwens (2003, S. 9-2).
TMC Total mass balance controller. Siehe van Griensven & Bauwens (2003, S. 9-3).
MNDEV Mittlere Abweichung:
P
M N DEV = ( xo − xs ) /n.
RE Relativer Fehler (%):
RE = 100 × M N DEV /x¯o .
97
TE Totaler Fehler.
P
P
T E = ( |xo − xs |) / |xo |.
R2 Bestimmtheitsmaß.
b Steigung des linearen Regressionsmodells.
n Gültige Wertepaare.
Mean Arithmetisches Mittel der simulierten und beobachteten Werte. Es werden nur die
paarweise gültigen Fälle zur Mittellung berücksichtigt.
sd Standardabweichung der simulierten und beobachteten Werte. Nur paarweise gültige
Fälle werden berücksichtigt.
sf Schiefe der Verteilungen, geschätzt nach dem Momentverfahren.
Ex Exzess der Verteilung, geschätzt nach dem Momentverfahren.
A.5.2 Anwendung
Das Programm benötigt eine Eingabedatei im .csv Format. Die Datei muss 2 Spalten
besitzen, wobei in der ersten Spalte die beobachteten und in der zweiten Spalte die simulierten Werte stehen müssen. Trennzeichen ist entweder Komma (,) oder Leerzeichen ( )
und das Dezimalzeichen ist Punkt (.). Es muss eine einheitliche Fehlkennung (z.B.: -99.0)
für sämtliche fehlenden Werte eingetragen werden. Das Programm ist interaktiv, d.h. der
Dateiname und die Fehlkennung werden abgefragt.
A.6 Manual for hru2map
A.6.1 Purpose of hru2map
The program hru2map re-assigns the output of the SWAT2000 model (Neitsch et al.,
2002) on the basis of Hydrological Response Units (HRU) to a raster map for a given
date. Consequently, any of the 58 simulated output variables within the SWAT output
file basins.sbs can be converted into a gridded map and visualized with an appropriate
GIS.
A.6.2 History of hru2map
The program hru2map bases on the idea of Haverkamp (2000) and Haverkamp et al. (2005),
who developed a tool named OUTGRASS. OUTGRASS was coupled to the GRASS GIS version
4 and hinged on the output format of SWAT99.2. Since both versions of programs are
outdated, the program hru2map was re-written from scratch. It is loosely coupled with
the GIS ArcView3.2 (extension: spatial analyst) and SWAT2000. It only reads and writes
ARC/GRID ASCII maps as specified by ESRI, hence, other GISs which support this file
format can also be used to create the input maps for hru2map.
98
Figure A.1: Example map of re-assigned HRUs for the Dill catchment. White areas contain
no-data values as well as not re-assigned grid cells.
The program has been successfully used within the framework of the Collaborative
Research Center 299, where SWAT simulated long term actual evapotranspiration and
percolation out of the top soil, which was then gridded to a map with hru2map. Both the
evapotranspiration and the percolation maps were input for the model ATOMIS, which
simulates land use and land management specific heavy metal accumulation. Figure A.1
gives an example of a re-assigned HRU map for the Dill catchment in Germany.
99
A.6.3 How the program works
In the SWAT2000 model an HRU is defined as a unique combination of soil type and
land use class within one subbasin. HRUs do not necessarily form a single field (physical
discrete area); it is likely that grid cells containing the same HRU category value are
distributed throughout one subbasin. It is worth to note that the same soil type and land
use combination may exist in a different subbasin, but the resulting HRU will be classified
by SWAT2000 with a different category value.
The aggregation algorithms implemented in ArcSWAT follow the above mentioned definition of an HRU. Furthermore, a user defined threshold area for an HRU can be specified.
This determines the minimum fraction of an HRU within a subbasin which is required
to form an HRU. HRUs with a lesser fraction than the threshold will disappear. Hence a
part of the input information will be lost.
The HRU specific output values in basins.sbs are re-assigned to a grid map using the
information of 3 maps, the subbasins category map, the land use category map and the
soil type map, respectively. The re-assignment is achieved through a logical comparison
as given in Eq. A.2.
yi,j = xj,k
∀
lui,j = lak,j ∗ soli,j = slk,j
(A.2)
where y is the assigned value on the i-th grid cell within the j-th subbasin of the output
map, x is the value for the k-th HRU within the j-th subbasin, lu is the land use category
on the i-th cell within the j-th subbasin, la is the land use category for the k-th HRU
in the j-th subbasin, ∗ means logical AND, sol is the soil type category for the i-th cell
within the j-th subbasin and sl is the soil type category of the k-th HRU and the j-th
subbasin.
Since a part of grid cells can not be re-assigned with HRU specific values, grid cells containing no-data will appear within one subbasin (between 1 % and 30 %). The attributes
of nodata cells are asigned with area weighted averages as given in Eq. A.3 and Eq. A.4.
xj =
n
X
(wj,k xj,k ) 1/
n
X
wj,k
(A.3)
k=1
k=1
with x̄ weighted average of the variable in the j-th subbasin, w weight (cell counts) of
assigned values of the k-th HRU in the j-th subbasin, and x value of the k-th HRU within
the j-th suburban.
yi,j = xj
∀
yi,j = N U LL
(A.4)
A.6.4 How to use the program
The program requires the files as provided in Tab. A.4. The file names are not important
except of the files basins.cod, basins.sbs and *.hru. The land use look up table has
the form as specified below. It complies with the specification 2 given by Di Luzio et al.
2
NOTE: The character string of a land use class MUST be 4 characters long. Upper and lower case are
not distinguished.
100
Table A.4: Input files for hru2map.
File name
Explanation
watershed.asc
ARC/GRID ASCII formatted
map of the watershed (categorical)
landuse.asc
map of the land use (categorical)
soil.asc
soil map (categorical)
soil_look_up.dat ASCII formatted soil look up table
lu_look_up.dat
ASCII formatted land use look up table
basins.cod
SWAT configuration file
basins.sbs
SWAT output file for each HRU
*.hru
Each SWAT .HRU file of the project
(2001, p. 13).
"Value","Landuse"
10,WATR
1,AGRL
2,PAST
3,CVEG
4,FRSD
5,FRSE
6,URBN
The soil look up table is formatted accordingly3 .
"Value","Name"
173,Soil-173
177,Soil-177
185,Soil-185
188,Soil-188
202,Soil-202
225,Soil-225
235,Soil-235
260,Soil-260
An ARC/INFO ASCII grid map contains four header lines and the actual values for
each grid cell in a matrix format as given in the below example.
ncols
nrows
xllcorner
3
1515
1437
3432100
NOTE: The soil name may take up to 9 characters. It MUST NOT include any white spaces!
101
yllcorner
5603100
1 2 2 2 1 3 3 ...
1 1 2 2 1 1 3 ...
...
You can export the grid maps from within ArcView by doing the following steps:
1. Start ArcView3.2 and load the extension “Spatial Analyst”.
2. Choose the following from the pull-down-menu:
FILE | EXPORT DATA SOURCE
3. Use the following export filter type:
ASCII Raster
4. Find your grid map in your specific folder:
C:\my-project\watershed\grids\lusgr
C:\my-project\watershed\grids\solgr
C:\my-project\watershed\grids\water
5. Save the grid files in ARC/INFO ASCII grid format by providing new file names.
If you work with GRASS-GIS, you will be able to use the module r.out.arc to export
a grid map in binary form into the ARC/INFO ASCII format.
To run hru2map you need to copy the executable file and each file listed in Tab. A.4
in a new folder. Open a console under Windows, change to the new created folder and
run hru2map. The program is interactive (i.e. asks the user for input files) and produces
two output files. The first one is called SUMMARY.OUT, which provides summary statistics
of the percentage of re-assigned area within one subbasin and the area weighted average
of the given value. The latter one is used to fill no-data values within one subbasin. The
second output file is named hrumap.asc, which contains the re-assigned values in the
ARC/INFO ASCII format. For further GIS based analyzes, the output grid map can
be imported into ArcView by using the function FILE | IMPORT DATA SOURCE; under
GRASS-GIS the module r.in.arc can be used.
A.6.5 Technical limitations
The program is limited to grid files with a maximum of 2,000 rows. Each file name of an
input ARC/GRID ASCII map may contain up to 25 characters (no white spaces!). You
can find other constraints regarding the look up tables on page 100.
A.6.6 Information for programmers
The program hru2map was written in Fortran90. You can find self explanatory comments in the source code. Special attention was taken for modularity, speed optimization,
memory allocation, and readability of the source code. The single procedures were organized to form functional program units. Table A.5 summarizes the main program and
102
Procedure
main.f90
readhru.f90
readcod2000.f90
loopyear.f90
readsbs2000.f90
readmaps.f90
mapdisagg.f90
aveandfill.f90
writesummary.f90
writemap.f90
readlookup.f90
upcase.f90
Table A.5: Procedures of hru2map
Purpose
The main program. Does some user chats.
Gets the number of HRUs and
reads the first line of each .hru file.
Reads specifications from basins.cod (SWAT2000).
Returns maximum doy and days of February.
Reads values from basins.sbs (SWAT2000).
Reads the input maps.
Assigns HRU specific values to an output map.
Fills nodata values within subbasins with weighted averages.
Writes summary output to SUMMARY.OUT.
Writes output map in ARC/GRID format to hrumap.asc.
Reads values from look up tables.
Upcases the character strings for land use and soil type.
other procedures in the sequential order of the calling structure. The program consists of
two modules, which provide globally defined variables (Tab. A.6).
103
Table A.6: Modules and their globally defined variables.
Module / variable Data type
Description
basinsmod.f90
Declares variables for readsbs2000
value
nr1
nr2
nr3
flag1
flag2
flag3
rchfla
bsbfla
sbsfla
rchvar
bsbvar
sbsvar
global.f90
nhru
nbas
ncols
nrows
ntot
nodata
soilm
lum
sba
lo
up
gsoilid
gluid
basmnmxhru
cellbas
cellbasmx
hrucell
maps
maphead
valsbs
104
integer
ID of selected variable
Nr. of printed variables in file .sbs
Indicators for customized output
integer(:)
Contains IDs of output variables
character(:)
Contains names of output variables
Declares variables for various procedures
integer
int,parameter
integer(:)
integer(:,:)
character(:)
real(:)
Number of HRUs in project
Number of subbasins in project
Number of columns in ARC/GRID map
Number of rows in ARC/GRID map
Total number of cells (nrows * ncols)
Value for nodata in ARC/GRID map
Named constant for soil map
Named constant for land use map
Named constant for subbasins map
Named constant for lowest HRU in subbasin
Named constant for upper HRU in subbasin
Contains soil-id per HRU
Contais land-use-id per HRU
Contains min. and max. HRU per subbasin
Contains number of assigned cells per subbasin
Conains number of total cells per subbasin
Contains number of cells per HRU
Contains soil, land use and subbasins map
Contains header of ARC/GRID map
Values per HRU which will be assigned
Literaturverzeichnis
2000/60/EG: Richtlinie 2000/60/EG des europäischen Parlaments und des Rates vom 23.
Oktober 2000 zur Schaffung eines Ordnungsrahmens für Maßnahmen der Gemeinschaft
im Bereich der Wasserpolitik, ABl. EG, S. L 327/1–L 327/27, 2000.
AG Boden: Bodenkundliche Kartieranleitung, Schweizbart’sche Verlagsbuchhandlung,
Hannover, 4 edn., 1994.
Angstrøm, A.: Solar and terrestrial radiation, Q. J. R. Meteorol. Soc., 50, 121–125, 1924.
Arheimer, B. & Brandt, M.: Modelling nitrogen transport and retention in the catchments
of southern Sweden, Ambio, 27, 471–480, 1998.
Arnold, J., Srinivasan, R., Muttiah, R., & Williams, J.: Large area hydrologic modeling
and assessment - Part 1: Model development, J. Am. Water Resour. Assoc., 34, 73–89,
1998.
Arnold, J. G. & Fohrer, N.: SWAT2000: current capabilities and research opportunities
in applied watershed modelling, Hydrol. Process., 19, 563–572, doi:10.1002/hyp.5611,
2005.
Arnold, J. G., Williams, J. R., & Maidment, D.: Continuous-time water and sediment
routing model for large basins, J. Hydrol. Eng., 121, 171–183, 1995.
Beaujouan, V., Durand, P., & Ruiz, L.: Modelling the effect of the spatial distribution of
agricultural practices on nitrogen fluxes in rural catchments, Ecol. Model., 137, 93–105,
2001.
Behrendt, H., Huber, P., Kornmilch, M., Opitz, D., Schmoll, O., Scholz, G., & Uebe,
R.: Nutrient emissions into river basins of Germany, Nr. 23/00 in UBA-Texte, Berlin,
http://www.umweltbundesamt.de/search-public.php, 2000.
Bergström, S.: The HBV model, in: Singh (1995), S. 443–476, 1995.
Beven, K. J. & Binley, A. M.: The future of distributed models: model calibration and
uncertainty prediction, Hydrol. Process., 6, 279–298, 1992.
Beven, K. J. & Freer, J.: Equifinality, data assimilation, and uncertainty estimation in
mechanistic modelling of complex environmental systems using the GLUE methodology,
J. Hydrol., 249, 11–29, 2001.
105
Beven, K. J., Lamb, R., Quinn, P., Romanowicz, R., & Freer, J.: TOPMODEL, in: Singh
(1995), S. 627–668, 1995.
Bicknell, B. R., Imhoff, J. C., Kittle, J. L., Donigian, A. S., & Johanson, R.: Hydrological
Simulation Program–Fortran: User’s manual for version 11, Environmental Protection
Agency, National Exposure Research Laboratory, Athens, GA., EPA/600/R-97/080,
1997.
Birkinshaw, S. J. & Ewen, J.: Nitrogen transformation component for SHETRAN
catchment nitrate transport modelling, J. Hydrol., 230, 1–17, 2000.
Borah, D. K. & Bera, M.: Watershed-scale hydrologic and nonpoint-source pollution models: Review of mathematical bases, Trans. ASAE, 46, 1553–1566, 2003.
Bouraoui, F., Grizzetti, B., Granlund, K., Rekolainen, S., & Bidoglio, G.: Impact of climate change on the water cycle and nutrient losses in a Finnish catchment, Climatic
Change, 66, 109–126, 2004.
Breuer, L., Eckhardt, K., & Frede, H.-G.: Plant parameter values for models in temperate
climates, Ecol. Model., 169, 237–293, 2003.
Brown, L. C. & Barnwell, T. O.: The enhanced water quality models QUAL2E and
QUAL2E-UNCAS documentation and user manual, USEPA, Athens, GA, epa document epa/600/3-87/007 edn., 1987.
Chaplot, V., Saleh, A., Jaynes, D., & Arnold, J.: Predicting water, sediment and NO3-N
loads under scenarios of land-use and management practices in a flat watershed, Water
Air Soil Pollut., 154, 271–293, 2004.
Christiaens, K. & Feyen, J.: Use of sensitivity and uncertainty measures in distributed
hydrological modeling with an application to the MIKE SHE model, Water Resour.
Res., 38, 1169–1184, doi:10.1029/2001WR000478, 2002.
Conan, C., Bouraoui, F., Turpin, N., de Marsily, G., & Bidoglio, G.: Modeling flow and
nitrate fate at catchment scale in Brittany (France), J. Environ. Qual., 32, 2026–2032,
2003.
Cui, J., Li, C., Sun, G., & Trettin, C.: Linkage of MIKE SHE to Wetland-DNDC for
carbon budgeting and anaerobic biogeochemistry simulation, Biogeochemi., 72, 147–
167, doi:10.1007/s10533-004-0367-8, 2005.
de Wit, M. J. M.: Nutrient fluxes at the river basin scale. I: the PolFlow model, Hydrol.
Process., 15, 743–759, 2001.
Di Luzio, M., Srinivasan, R., & Arnold, J.: ArcView interface for SWAT2000. User’s
guide., USDA-ARS, Temple, TX, 2001.
Di Luzio, M., Srinivasan, R., & Arnold, J.: Integration of watershed tools and SWAT
model into BASINS, J. Am. Water Resour. Assoc., 38, 1127–1141, 2002.
106
DIN EN ISO 10304-1: Bestimmung der gelösten Anionen Fluorid, Chlorid, Nitrit, Orthophosphat, Bromid, Nitrat und Sulfat mittels Ionenchromatographie. Teil 1: Verfahren
für gering belastete Wässer, 1995.
Dreyhaupt, J., Hrsg.: Stickstoffmodellierung für Lysimeter des Parthe-Gebietes. Ergebnisse des Workshops “Stickstoffmodellierung“ vom 8.6.1999 bis 10.6.1999 am UFZ LeipzigHalle, Bd. 17/2000 von UFZ Bericht, Leipzig, 2000.
Duan, Q., Sorooshian, S., & Gupta, V.: Effective and efficient global optimisation for
conceptual rainfall-runoff models, Water Resour. Res., 28, 1015–1031, 1992.
DVWK, Hrsg.: Ermittlung der Verdunstung von Land- und Wasserflächen, Bd. 238 von
Merkblätter zur Wasserwirtschaft, Wirtschafts- und Verlagsgesellschaft Gas und Wasser, Bonn, 1996.
Eckhardt, K. & Arnold, J.: Automatic calibration of a distributed catchment model, J.
Hydrol., 251, 103–109, 2001.
Eckhardt, K., Haverkamp, S., Fohrer, N., & Frede, H.-G.: SWAT-G, a version of SWAT99.2
modified for application to low mountain range catchments, Phys. Chem. Earth, 27,
641–644, 2002.
Eckhardt, K., Fohrer, N., & Frede, H.-G.: Automatic model calibration, Hydrol. Process.,
19, 651–658, doi:10.1002/hyp.5613, 2005.
Frede, H.-G. & Bach, M.: Perspectives for peripheral regions, Z. Kulturtech. Landentw.,
40, 193–196, in German, 1999.
Friedrich, C. & Lenhart, T.: Generierung und Datenbestand Bodenkarte, ILR-Manuskript,
ohne Jahr.
Fröhlich, H., Opp, C., & Scholten, T.: Periglacial layer or anthropogenic layer? – A small
scale analysis of their spatial distribution under forest, Z. Geomorph. N. F., 139, 203–
219, 2005.
Fürst, J.: GIS in Hydrologie und Wasserwirtschaft, Wichmann, Heidelberg, 2004.
Gauger, T., Köble, R., Spranger, T., Bleeker, A., & Draaijers, G.: Deposition of sulphur
and nitrogen in Germany, Water Air Soil Pollut., 1, 353–373, 2001.
Gebel, M.: Entwicklung und Anwendung des Modells N-BILANZ zur Quantifizierung von
Stickstoffeinträgen in mesoskaligen Flusseinzugsgebieten, Diss. , TU Dresden, 2000.
Grayson, R. & Blöschl, G.: Spatial modelling of catchment dynamics, in: Spatial patterns
in catchment hydrology, hrsg. von Grayson, R. & Blöschl, G., S. 51–81, Cambridge
University press, Cambridge, 2000.
Grayson, R. B., Gippel, C. J., Finlayson, B. J., & Hart, B. T.: Catchment-wide impacts
on water quality: the use of ’snapshot’ sampling during stable flow, J. Hydrol., 199,
121–134, 1997.
107
Grizetti, B., Bouraoui, F., de Marsily, G., & Bidoglio, G.: A statistical method for source apportionment of riverine nitrogen loads, J. Hydrol., 304, 302–315, doi:10.1016/j.
jhydrol.2004.07.036, 2005.
Grizzetti, B., Bouraoui, F., Granlund, K., Rekolainen, S., & Bidoglio, G.: Modelling diffuse
emission and retention of nutrients in the Vantaanjoki watershed (Finland) using the
SWAT model, Ecol. Model., 169, 25–38, 2003.
Haferkorn, U.: Größen des Wasserhaushaltes verschiedener Böden unter landwirtschaftlicher Nutzung im klimatischen Grenzraum des Mitteldeutschen Trockengebietes - Ergebnisse der Lysimeterstation Brandis, Diss. , Univ. Göttingen, Göttingen, 2000.
Haverkamp, S.: Methodenentwicklung zur GIS-gestützten Modellierung des Landschaftswasserhaushaltes, Diss. , Justus-Liebig-Universität Gießen, Gießen, 2000.
Haverkamp, S., Fohrer, N., & Frede, H.-G.: Assessment of the effect of land use patterns
on hydrologic landscape functions: a comprehensive GIS-based tool to minimize model
uncertainty resulting from spatial aggregation, Hydrol. Process., 19, 715–727, doi:10.
1002/hyp.5626, 2005.
Hessisches Landesamt für Umwelt und Geologie, Hrsg.: Digitale Bodenflächendaten von
Hessen 1 : 50 000 (BFD50), Version: 1, 2000.
Hogue, T. S., Sooroshian, S., Gupta, H., Holz, A., & Braatz, D.: A multistep automatic
calibration scheme for river forecasting models, J. Hydromet., 1, 524–542, 2000.
Horn, A. L., Rueda, F. J., Hörmann, G., & Fohrer, N.: Implementing river water quality
modelling issues in mesoscale watershed models for water policy demands—an overview
on current concepts, deficits, and future tasks, Phys. Chem. Earth, 29, 725–737, doi:
10.1016/j.pce.2004.05.001, 2004.
Huisman, J. A., Breuer, L., Eckhardt, K., & Frede, H.-G.: Spatial consistency of automatically calibrated SWAT simulations in the Dill catchment and three of its subcatchments, in: 2003 International SWAT Conference, hrsg. von Srinivasan, R., Jacobs,
J. H., & Jensen, R., Bd. 266 von TWRI Technical Report, S. 168–173, Bari, 2004.
Huisman, J. A., Pohlert, T., Breuer, L., & Frede, H.-G.: The power of multi-objective
calibration: two case studies with SWAT, in: Srinivasan et al. (2005), S. 200–209, http:
//www.brc.tamus.edu/swat/index.html, visited on 4 May 2006, 2005.
Hydro Agri Dülmen, Hrsg.: Faustzahlen für Landwirtschaft und Gartenbau, Dülmen, 1993.
IPCC, Hrsg.: Greenhouse Gas Inventory Reference Manual. Revised 1996 IPCC Guidelines
for National Greenhouse Gas Inventories, Bd. 3, Reference Manual, Kap. 4 Agriculture,
IPCC WGI Technical Support Unit, Bracknell, UK, www.ipcc-nggip.iges.or.jp/
public/gl/invs6.htm, visited on 14 Nov. 2006, 1997.
108
Jarvie, H. P., Wade, A. J., Butterfield, D., Whitehead, P. G., Tindall, C. I., Virtue, W. A.,
Dryburgh, W., & McGraw, A.: Modelling nitrogen dynamics and distributions in the
River Tweed, Scottland: and application of the INCA model, Hydrol. Earth. Syst. Sci.,
6, 433–453, 2002.
Jones, C. A. & Kiniry, J. R.: CERES-Maize: A simulation model of maize growth and
development, Techn. Ber., Texas A&M University Press, College Station, TX, 1986.
Jungkunst, H. F., Freibauer, A., Neufeldt, H., & Bareth, G.: Nitrous oxide emissions from
agricultural land use in Germany—a synthesis of available annual field data, J. Plant
Nutr. Soil Sci., 169, 341–351, doi:10.1002/jpln.200521954, 2006.
Kaviany, E.: Zur Hydrogeologie im Niederschlagsgebiet der Dill (Hessen), Diss. , JustusLiebig-Universität Gießen, 1978.
Kersebaum, K. C.: Application of a simple management model to simulate water and
nitrogen dynamics, Ecol. Model., 81, 145–156, 1995.
Klemeš, V.: Operational testing of hydrological simulation models, Hydrol. Sci. J., 31,
13–24, 1986.
Knappe, S., Haferkorn, U., & Meissner, R.: Influence of different agricultural management
systems on nitrogen leaching: results of lysimeter studies, J. Plant Nutr. Soil Sci., 165,
73–77, 2002.
Kronvang, B., Hezlar, J., Boers, P., Jansen, J. P., Behrendt, H., Anderson, T., Arheimer,
B., Venohr, M., & Hoffmann, C. C.: Nutrient retention handbook. Software manual
for EUROHARP-NUTRET and scientific review on nutrient retention, Nr. 9-2004 in
EUROHARP report, Oslo, www.euroharp.org, 2004.
Krysanova, V. & Haberlandt, U.: Assessment of nitrogen leaching from arable land in
large river basins. Part I. Simulation experiment using a process-based model, Ecol.
Model., 150, 255–275, 2002.
Kuczera, G. & Franks, S. W.: Testing hydrologic models: fortification or falsification, in:
Singh & Frevert (2002), S. 141–186, 2002.
Legates, D. R. & McCabe, G. J.: Evaluating the use of ”goodness-of-fit”measures in
hydrologic and hydroclimatic model validation, Water Resour. Res., 35, 233–241, 1999.
Leirós, M. C., Trasar-Cepeda, C., Seoane, S., & Gil-Sotres, F.: Dependence of mineralization of soil organic matter on temperature and moisture, Soil Biol. Biochem., 31,
327–335, 1999.
Lenhart, T.: Modellierung des Stoffhaushaltes in mesoskaligen Einzugsgebieten, Diss. ,
Justus-Liebig-Universität Gießen, Gießen, 2003.
Lenhart, T., Eckhardt, K., Fohrer, N., & Frede, H. G.: Comparison of two different approaches of sensitivity analysis, Phys. Chem. Earth, 27, 645–654, 2002.
109
Lenhart, T., Fohrer, N., & Frede, H.-G.: Effects of land use changes on the nutrient balance
in mesoscale catchments, Phys. Chem. Earth, 28, 1301–1309, doi:10.1016/j.pce.2003.09.
006, 2003.
Li, C., Frolking, S., & Frolking, T. A.: A model of nitrous oxide evolution from soil driven
by rainfall events: 1. Model structure and sensitivity, J. Geophys. Res., 97, 9759–9776,
1992.
Li, C., Aber, J., Stange, F., Butterbach-Bahl, K., & Papen, H.: A process-oriented model
of N2 O and NO emissions from forest soils: 1. Model development, J. Geophys. Res.,
105, 4369–4384, 2000.
Li, X., Ambrose, R. B., & Araujo, R.: Modeling mineral nitrogen export from a forest
terrestrial ecosystem to streams, Trans. ASAE, 47, 727–739, 2004.
Liu, S., Tucker, P., Mansell, M., & Hursthouse, A.: Development and application of a
catchment scale diffuse nitrate modelling tool, Hydrol. Process., 19, 2625–2639, doi:
10.1002/hyp.5678, 2005.
Ludwig, B., Wolf, I., & Teepe, R.: Contribution of nitrification and denitrification to the
emission of N2 O in a freeze-thaw event in an agricultural soil, J. Plant Nutr. Soil Sci.,
167, 678–684, doi:10.1002/jpln.200421462, 2004.
Marchetti, R., Donatelli, M., & Spallacci, P.: Testing denitrification functions of dynamic
crop models, J. Environ. Qual., 26, 394–401, 1997.
Meek, D. W. & Hatfield, J. L.: Data quality checking for single station meteorological
databases, Agric. Forest. Meteorol., 69, 85–109, 1994.
Muleta, M. K. & Nicklow, J. W.: Sensitivity and uncertainty analysis coupled with automatic calibration for a distributed watershed model, J. Hydrol., 306, 127–145, doi:
10.1016/j.jhydrol.2004.09.005, 2005.
Müller, C., Martin, M., Stevens, R. J., Laughlin, R., Kammann, C., Ottow, J. C. G., &
Jäger, H.-J.: Processes leading to N2 O emissions in grassland soil during freezing and
thawing, Soil Bio. Biochem., 34, 1325–1331, 2002.
Nash, J. E. & Sutcliffe, J. V.: River flow forecasting through conceptual models. Part 1.
A discussion of principles, J. Hydrol., 10, 282–290, 1970.
Neitsch, S. L., Arnold, J. G., Kinry, J. R., & Williams, J. R.: Soil and water assessment
tool. Version 2000. Theoretical documentation, USDA-ARS, Temple, TX, 2002.
Niklińska, M., Maryańska, M., & Laskowski, R.: Effect of temperature on humus respiration rate and nitrogen mineralization: Implications for global climate change, Biogeochemi., 44, 239–257, 1999.
Nöhles, I.: Landnutzungsklassifikation mit multitemporalen Landsat TM-Szenen in einer
kleinstrukturierten Agrarregion, Diss. , Justus-Liebig-Universität Gießen, Gießen, 2000.
110
Pohlert, T.: Manual for hru2map, Institute for Landscape Ecology and Resources Management, Gießen, www.uni-giessen.de/ilr/frede/swat/swat.html, 2005.
Pohlert, T., Breuer, L., Huisman, J. A., & Frede, H.-G.: Integration of a detailed nitrogen
balance model into a mesoscale water balance model for river water quality predictions,
in: Sharing a common vision of our water resources. 6th International Conference.
European Water Resources Association. 7 - 10 September 2005. Palais de l’Europe.
Menton (France), Paper Nr. EWRA151, EWRA, 2005a.
Pohlert, T., Huisman, J. A., Breuer, L., & Frede, H.-G.: Modeling of point and non-point
source pollution of nitrate with SWAT in the river Dill, Germany, Adv. Geosci., 5, 7–12,
2005b.
Pohlert, T., Huisman, J. A., Breuer, L., & Frede, H.-G.: Evaluation of the soil nitrogen
balance model in SWAT with lysimeter data, in: Srinivasan et al. (2005), S. 498–508,
http://www.brc.tamus.edu/swat/index.html, visited on 4 May 2006, 2005c.
Pohlert, T., Breuer, L., Huisman, J. A., & Frede, H.-G.: Integration of a detailed biogeochemical model into SWAT for improved nitrogen predictions – model development,
sensitivity and uncertainty analysis, Ecol. Model., in review, 2006a.
Pohlert, T., Breuer, L., Huisman, J. A., & Frede, H.-G.: Assessing the model performance
of an integrated hydrological and biogeochemical model for discharge and nitrate load
predictions, Hydrol. Earth Syst. Sci. Discuss., 3, 2813–2851, 2006b.
R Development Core Team: R: A language and environment for statistical computing, R
Foundation for Statistical Computing, Vienna, Austria, http://www.R-project.org,
2004.
Refsgaard, J. C.: Parameterisation, calibration and validation of distributed hydrological
models, J. Hydrol., 198, 69–97, 1997.
Refsgaard, J. C. & Henriksen, H. J.: Modelling guidelines — terminology and guiding
principles, Adv. Water R., 27, 71–82, doi:10.1016/j.advwatres.2003.08.006, 2004.
Refsgaard, J. C. & Storm, B.: MIKE SHE, in: Singh (1995), S. 809–846, 1995.
Richter, D.: Ergebnisse methodischer Untersuchungen zur Korrektur des systematischen
Meßfehlers des Hellmann-Niederschlagsmessers, Bd. 194 von Berichte DWD, Selbstverlag DWD, Offenbach, 1995.
Rodrigo, A., Recous, S., Neel, C., & Mary, B.: Modelling temperature and moisture effects
on C-N transformations in soils: comparison of nine models, Ecol. Model., 102, 325–339,
1997.
Santhi, C., Arnold, J., Williams, J., Dugas, W., Srinivasan, R., & Hauck, L.: Validation of
the SWAT model on a large river basin with point and nonpoint sources, J. Am. Water
Resour. Assoc., 37, 1169–1188, 2001a.
111
Santhi, C., Arnold, J., Williams, J., Hauck, L., & Dugas, W.: Application of a watershed
model to evaluate management effects on point and nonpoint source pollution, Trans.
ASAE, 44, 1559–1570, 2001b.
Sauer, D.: Genese, Verbreitung und Eigenschaften periglaziärer Lagen im Rheinischen
Schiefergebirge - anhand von Beispielen aus Westerwald, Hunsrück und Eifel , Diss. ,
Justus-Liebig-Universität Gießen, Gießen, 2002.
Sieber, A. & Uhlenbrook, S.: Sensitivity analyses of a distributed catchment model to
verify the model structure, J. Hydrol., 310, 216–235, doi:10.1016/j.jhydrol.2005.01.004,
2005.
Singh, V. P., Hrsg.: Computer models of watershed hydrology, Water Resources Publications, Highlands Ranch, CO, 1995.
Singh, V. P. & Frevert, D. K., Hrsg.: Mathematical models of large watershed hydrology,
Water Resources Publication, Highlands Ranch, CO, 2002.
Sivapalan, M., Viney, N. R., & Zammit, C.: LASCAM: Large scale catchment model, in:
Singh & Frevert (2002), Kap. 14, S. 579–648, 2002.
Sivapalan, M., Takeuchi, K., Franks, S. W., Gupta, V. K., Karambiri, H., Lakshmi, V.,
Liang, X., McDonnell, J. J., Mendiondo, E. M., O’Connell, P. E., Oki, T., Pomeroy,
J. W., Schertzer, D., Uhlenbrook, S., & Zehe, E.: IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012: Shaping an exciting future for the hydrological sciences,
Hydrol. Sci. J., 48, 857–880, 2003.
Sophocleous, M. & Perkins, S.: Methodology and application of combined watershed and
ground-water models in Kansas, J. Hydrol., 236, 185–201, 2000.
Srinivasan, R., Jacobs, J., Day, D., & Abbaspour, K., Hrsg.: 3rd International SWAT
Conference, 2005, Zurich, http://www.brc.tamus.edu/swat/index.html, visited on
4 May 2006, 2005.
Stange, F.: Entwicklung und Anwendung eines prozeßorientierten Modells zur Beschreibung der N2 O und NO-Emissionen aus Böden temperater Wälder, Diss. , Univ. Freiburg, Garmisch-Partenkirchen, 2001.
Stöckle, C. O. & Campbell, G. S.: Simulation of crop response to water and nitrogen: an
application example using spring wheat, Trans. ASAE, 32, 66–74, 1989.
Stöckle, C. O., Donatelli, M., & Nelson, R.: CropSyst, a cropping systems simulation
model, Europ. J. Agronomy, 18, 289–307, 2003.
Supit, I., Hooijer, A. A., & van Diepen, C. A.: System description of the WOFOST 6.0
crop simulation model implemented in CGMS, Techn. Ber., Joint Research Centre EC,
1994.
van Griensven, A. & Bauwens, W.: Multiobjective autocalibration for semidistributed
water quality models, Water Resour. Res., 39, 1348, doi:10.1029/2003WR002284, 2003.
112
Viney, N. R., Sivapalan, M., & Deeley, D.: A conceptual model of nutrient mobilisation
and transport applicable at large catchment scales, J. Hydrol., 240, 23–44, 2000.
Vrugt, J. A., Gupta, H. V., Bastidas, L. A., Bouten, W., & Sorooshian, S.: Effective and
efficient algorithm for multiobjective optimization of hydrologic models, Water Resour.
Res., 39, 1214, doi:10.1029/2002WR001746, 2003.
Wade, A., Durand, P., Beaujouan, V., Wessel, W., Raat, K., Whitehead, P., Butterfield,
D., Rankinen, K., & Lepisto, A.: A nitrogen model for European catchments: INCA,
new model structure and equations, Hydrol. Earth Sys. Sci., 6, 559–582, 2002.
Weinmann, B., Schroers, J. O., Sheridan, P., & Kuhlmann, F.: Modelling the CAP reform
at the regional level with ProLand., in: Proc. European Association of Agricultural
Economists (EAAE) Congress 2005 at the Royal Veterinary and Agricultural University
in Copenhagen, Denmark, EAAS, Copenhagen, 2005.
Whitehead, P. G., Wilson, E. J., & Butterfield, D.: A semi-distributed Integrated Nitrogen
model for multiple source assessment in Catchments (INCA): Part 1. - model structure
and process equations, Sci. Total Environ., 210/211, 547–558, 1998a.
Whitehead, P. G., Wilson, E. J., Butterfield, D., & Seed, K.: A semi-distributed integrated
flow and nitrogen model for multiple source assessment in catchments (INCA): Part 2. application to large river basins in south Wales and eastern England, Sci. Total Environ.,
210/211, 559–583, 1998b.
Williams, J. R., Jones, C. A., & Dyke, P. T.: A modeling approach to determining the
relationship between erosion and soil productivity, Trans. ASAE, 27, 129–144, 1984.
Wilmott, C. J., Ackleson, S. G., Davis, R. E., Feddema, J. J., Klink, K. M., Legates,
D. R., O’Donnel, J., & Rowe, C. M.: Statistics for the evaluation and comparison of
models, J. Geophys. Res., 90, 8995–9005, 1985.
Zhang, Y., Li, C., Zhou, X., & Moore, B.: A simulation model linking crop growth and
soil biogeochemistry for sustainable agriculture, Ecol. Model., 151, 75–108, 2002.
113

Modellierung des Stickstoffhaushaltes für den

Transcrição

Documentos relacionados

Fische beizen - Angelverein Malente

Karte doppelseitig.cdr

65 70 97 Aalborg Jubilaeums Aquavit

REZEPTSAMMLUNG

Minerale führenden Goldseifen aus Niederbayern, Deutschland

Finkenwerder Scholle mit karamellisierten Kartoffeln

Die anaerobe Bebrütung zur Bestimmung des N

blaues Poloshirt / SWAT Basic-Instructor - Self

Avoca Caustic Applicators 75 % and 95 % Silver Nitrate Caustic

Stickstoff-Federsysteme / Manifold systems